Abstract
Research in the field of representative bureaucracy provides evidence that the presence of minority and female bureaucrats can improve outcomes for minority and female clients of an agency. However, the previous work has almost exclusively connected the benefits of representation to outcomes directly linked to the primary purpose of the organization. We contend that there may be additional benefits linked to bureaucratic representation that are beyond the mission of the agency. To test this contention, we examine whether the presence of minority and female teachers can improve outcomes, beyond educational measures, for female students. Specifically, using data from Georgia public schools, we test whether the presence of minority and female teachers lowers teen pregnancy rates. We find that as the representation of African-American teachers reaches the tipping point (around 20%) we see a significant reduction in the African-American teen pregnancy rate.
Research in the field of representative bureaucracy provides evidence that the presence of minority and female bureaucrats can improve outcomes for minority and female clients of an agency. However, the previous work has almost exclusively connected the benefits of representation to outcomes directly linked to the primary mission or function of the organization. For example, scholars find that female Child Support Enforcement supervisors increase child support payments to custodial mothers (Wilkins and Keiser 2006). However, we contend that there may be additional pro-social benefits linked to bureaucratic representation that are beyond the mission of the agency.
Numerous studies (Hindera 1993; Keiser et al. 2002; Meier and Nicholson-Crotty 2006; Meier and Stewart 1992; Meier, Stewart and England 1989; Selden 1997; Theobald and Haider-Markel 2009; Wilkins and Keiser 2006) have concluded that the presence of minority and female employees can reduce the disparate treatment minority and female clients have historically received from various public bureaucracies. The reduction in disparate outcomes may be the result of minorities and women being present in the organization or the direct actions of these minority or female bureaucrats to advocate for clients who share their demographic characteristics. We expand on this work by examining whether the presence of minority and female bureaucrats can provide benefits beyond the organization, to society as a whole. Although we expect our analysis to offer theoretical developments in the literature on representative bureaucracy, the policy implications of our findings are important in the research examining ways to reduce unintended pregnancies among teenagers.
The prevalence of unintended pregnancies among adolescents in the United States is high and concerning. Among fully industrialized countries, the United States has the highest rate of teen pregnancy (Hoffman 2006). The rate of teen pregnancy in the United States is almost twice that of Great Britain, four times that of France and Germany, and over 10 times that of Japan (Hoffman 2006). The consequences of these behaviors on future health and economic outcomes can be devastating. The US taxpayers spend approximately $9.1 billion a year on teen pregnancy (Hoffman 2006). Thus, much research has been devoted to understanding the determinants of these behaviors to inform prevention and policy interventions (Gruber 2001). Researchers have examined numerous individual, family, and community determinants of adolescent sexual behavior. In addition, several studies have highlighted the need for sex education programs to be culturally relevant and sensitive for African-American and Latina students (Crump et al. 1999; Dixon, Schoonmaker, and Philliper 2000; Horner et al. 2008; Rosenthal, Lewis, and Cohen 1996; Tabi 2002; Taylor and Ward 1991). We contend that research in the field of representative bureaucracy offers a potentially new factor to examine.
To test this contention, we examine whether the presence of minority and female teachers can improve outcomes, beyond the classroom, for female students. Specifically, using data from Georgia public schools we test whether the presence of minority and female teachers lowers teen pregnancy rates.1 We expect that the impact of this representation will vary by the race of the student, with coethnic representation being most important for improving outcomes for African-American and Latina students.
Theory of Representative Bureaucracy
The theory of representative bureaucracy concerns how the demographic characteristics of bureaucrats affect the distribution of outputs to clients who share these demographic characteristics. Mosher (1968) distinguishes between two forms of representation—passive and active. A bureaucracy is passively representative, in Mosher’s terms, to the extent that it employs minorities and women in numbers proportionate to their shares of the population, or at least proportionate to those parts of the population with the qualifications necessary for employment. Studies of passive representation take one of two approaches. The first approach examines whether the composition of the bureaucracy mirrors the demographic composition of the general population (e.g., Dometrius and Sigelman 1984). The other approach identifies factors that increase or decrease the representation of women and minorities in the bureaucracy (Kellough 1990; Naff and Crum 2000; Riccucci and Saidel 1997).
Research on representative bureaucracy seldom examines how such passive representation may produce policy outputs. However, increasingly researchers understand that passive representation can influence outputs that are consistent with the interests of members of specific groups. The relationship between passive representation and policy outputs may be the indirect effect of minority bureaucrats inducing changes in the behavior of citizens rather than through the bureaucrat actively representing the interests of any group (Lim 2006). A growing body of research focuses on symbolic representation and argues that the presence of minority representatives,2 in and of itself, can change the behavior of the clients and their perceptions about the legitimacy of government. Symbolic representation does not require any purposeful action by the person holding the position—“being there” is enough to change the behavior of minority or female clients by making the agencies’ services more attractive to these clients (Meier and Nicholson-Crotty 2006; Theobald and Haider-Markel 2009). Researchers have found, for example, that women are more likely to report sexual assault to a police department where women are well represented among the officers (Meier and Nicholson-Crotty 2006). This suggests that the presence of female officers changes the behavior of females in the population, without any evidence that the female officers behave differently than their male counterparts. Similarly, Theobald and Haider-Markel (2009) find that African-American citizens are more likely to perceive police actions as being legitimate if there are African-American officers present. The same relationship holds for white citizens and white officers.
The second type of representation, specified by Mosher, occurs when public employees “press for the interests and desires of those whom [they] are presumed to represent” (1968). Mosher referred to this concept as active representation. It suggests that bureaucrats will act to see that the interests of individuals with whom they share an identity are appropriately considered as decisions affecting public policies and programs are made. In short, Mosher suggests that passive representation could (although it would not necessarily) lead to active representation. Scholars argue that a prerequisite for active representation is that bureaucrats share the values of those whom they are presumed to represent (Meier 1975; Meier and Nigro 1976; Saltzstein 1979). Researchers contend that it is the differences in attitudes and values and how the differences influence the behavior of the bureaucrat that provide the link between passive and active representation (Brown and Harris 1993, 2001; Brudney, Hebert, and Wright 2000; Dolan 2000, 2002).
As discussed earlier, several studies (Hindera 1993; Keiser et al. 2002; Meier and Nicholson-Crotty 2006; Meier and Stewart 1992; Meier, Stewart and England 1989; Selden 1997; Wilkins and Keiser 2006) conclude that minority and female bureaucrats implement policies or use their discretion to improve policy outcomes for minority clients. A large part of this work examines the link between passive and active representation in the context of public schools. Findings from these studies demonstrate that representation in public schools has been shown to relate to higher test scores for minority students (Meier, Wrinkle, and Polinard 1999), higher academic achievement for female students (Keiser et al. 2002), and fewer students dropping out of school (Pitts 2005). However, this work has been criticized because of the difficulty in knowing whether the effects attributed to active representation are actually just correlates of passive representation (Lim 2006).
There are several ways that the results we attribute to active representation may, in fact, be correlates of passive representation. First, there is growing evidence that symbolic representation can influence outputs indirectly by inducing changes in the attitudes and/or behavior of citizens rather than through the bureaucrat actively representing the interests of any group (Lim 2006; Meier and Nicholson-Crotty 2006; Pitkin 1967; Theobald and Hader-Markel 2009). A second possible causal mechanism is that the bureaucrat (e.g., a teacher) serves as a role model for the client (student) and stimulates productive changes in these clients. A recent study (Beaman et al. 2012) conducted in 495 villages in the state of West Bengal in India shows how young girls and their families are directly affected by powerful female role models. In 1993, this region started a quota that mandated a certain number of women to serve in government positions. Since then, social expectations for girls significantly changed. In the areas with female leaders, both the girls and their parents believed that girls deserved the same educational attainment as boys. In villages that did not have this quota, girls and their parents were significantly less likely to have aspirations of education for girls. These role models may or may not ever interact with the girls and their parents, but their presence produced positive benefits. Third, the presence of minority and female bureaucrats may lead to changes in the behavior of majority bureaucrats. Finally, the increases in representation may shift the policies and/or priorities of the organization. Given these alternative mechanisms, the benefits found in much of the research could be attributed to the direct actions of the bureaucrat or may be correlates of passive representation. We recognize that in many cases one or more of these causal mechanisms may be at work.
Regardless of whether you are testing the correlates of passive representation or the provision of active representation, it is important to understand the conditions under which you expect representation to alter outcomes for a particular clientele of an agency. Keiser et al. (2002) identified two necessary, but not always sufficient, conditions that must exist for sex to facilitate the transformation of passive representation to active representation. First, the policy area must be salient to the demographic group, including the clients, in question (Keiser et al. 2002; Meier 1993; Selden 1997). The second condition is that the policy area must be one in which bureaucrats have, or are at least believed by the client to have, discretion (Meier 1993). Although these conditions were theorized regarding the provision of active representation, we contend that, with limited revision, they must also be met for clients to experience the benefits of passive representation.
Research on representative bureaucracy demonstrates the role that institutional and contextual factors play in the link between passive and active representation (Keiser et al. 2002). One of the institutional variables considered is the presence of a critical mass of minority bureaucrats. Researchers argue that a critical mass may be needed for minorities and women to take an advocacy role (Kanter 1977; Meier 1993; Thompson 1978). In her work on business organizations, Kanter (1977) argued that when women make up less than 15% of the organization’s membership, they are merely “tokens” and lack the power and position to facilitate change within the organization. These “tokens” often try to minimize differences and keep a low profile. Although Kanter argues that 15% of the organization’s membership represents a critical mass, the necessary percentage is likely to vary by context. For example, in earlier work, Meier (1993) tests if a numerical threshold is needed for Latina principals in a school district to have a positive impact on the educational outcomes of Latina students. He finds that the critical mass for limiting corporal punishment is 24%, whereas limiting out-of-school suspension, offering alternative education assignments, and reducing expulsions all require 26% of the principals to be Latina. Given this evidence, it may not be until the minority or female presence in the organization reaches a critical mass that differences in outcomes are observed.
All of the previous work on the correlates of passive representation or active representation examines the impact of representation on outputs directly linked to the mission of the organization. However, we argue that the presence of minority and female bureaucrats, in some settings, can produce pro-social benefits for minority and female clients not directly tied to the mission of the agency. In other words, the benefits created by passive or active representation may extend beyond the mission and primary function of the agency the bureaucrat works in. This leads us to hypothesize that female and minority bureaucrats will improve outputs, tied and not tied to the mission of the organization, for clients who share their identity, either via race or gender, through their presence or direct actions. In addition, we posit that a critical mass of minority or female bureaucrats within the organization is necessary to see the positive effects associated with increased passive representation.
Empirical Case: Minority Teachers and Teen Pregnancy Rates
The teen pregnancy statistics are alarming. Thirty-four percent of girls get pregnant at least once before age 20, resulting in more than 820,000 teen pregnancies a year (Hoffman 2006). Less than one third of teen mothers finish high school and only 2% graduate from college (Perper, Peterson, and Manlove 2010). The child of a teen mother also faces serious challenges. Babies born to teenagers are more at risk for premature birth and low birth weight, raising the possibility of infant death, blindness, deafness, chronic respiratory problems, mental retardation, mental illness, cerebral palsy, dyslexia, and hyperactivity (Corcoran 1998; Ventura et al. 2001). Children born to teenagers are also at greater risk of neglect and abuse, have worse educational outcomes, and are less likely to receive adequate health care than children born to women over 20 years of age (Furstenberg, Brooks-Gunn, and Chase-Lansdale 1989). Sons of teen mothers are more likely to serve time in prison, and daughters of teen mothers are more likely to become teen mothers themselves (Perper, Peterson, and Manlove 2010). Large racial disparities in teen pregnancy exist. A 2010 report from the Guttmacher Institute indicates that the African-American teenage birth rate in 2006 was 126 per 1,000 women compared to 44 per 1,000 women for whites (Kost, Henshaw, and Carlin 2010). Even controlling for socioeconomic status, rates of teen pregnancy are significantly higher for African-Americans than any other group.
Numerous factors are understood to contribute to teen pregnancy rates; these factors operate at the individual, familial, and community levels (Danzinger 1995; Kirby 2001; Ladner 1987; Langille 2007). At the individual level, a teenager’s level of knowledge, attitudes toward sexual behavior, beliefs, expectations regarding the future, and use of alcohol and drugs are correlated with their decision to engage in sexual activity and the use of contraceptives (CDC 2008; Langille 2007). It is important to note that some of these issues are influenced by the race and/or ethnicity of the young woman. Minority teenagers are at greater risk of becoming pregnant due to a lack of knowledge and contraceptive use (CDC 2008). Twice as many white students than African-Americans or Latinas reported using birth control pills before their last sexual encounter (CDC 2008). Familial relationships also play a critical role—both intrafamilial and extrafamilial (Langille 2007). Research has found that the family structure, parent–child relationship, and socioeconomic status of the family all influence the sexual behaviors of teenagers (CDC 2008; Danzinger 1995; Kirby 2001; Ladner 1987; Langille 2007). Similarly, peer pressure, the level and type of sex education, and quality of health services are shown to be important to the reproductive health of teenage girls (CDC 2008; Kirby 2001; Langille 2007). The norms and values of the community the young woman belongs to will also influence her sexual behaviors. Recent research examines the importance of these norms and values in designing programs targeted at reducing teen pregnancy rates, especially for African-Americans and Latinas (Tabi 2002).
Evaluations of current pregnancy prevention programs directed toward African-American teenagers do not find strong and consistent results in decreasing teenage pregnancy (Tabi 2002). Tabi argues that the failure of these programs is due to the fact that they do not present culturally appropriate information to African-American teenagers. Often the information provided is not easily understood or meaningful to African-American teenagers (Tabi 2002). In response to the criticisms of traditional sex education programs, programs designed to specifically address the concerns of the African-American community have been introduced. For example, “A Journey toward Womanhood” is a 13-week pregnancy prevention program specifically designed from an Afrocentric approach. This program was created under the premise that pregnancy prevention programs appear to be more effective when designed in a more culturally sensitive manner (Dixon, Schoonmaker, and Philliper 2000). The program begins with a broad exploration of self through researching African-American heritage and culture, and in the final weeks, participants learn the importance of “Inner Health for Outer Beauty” as well as “The Tools for Survival.” The study compared sexual behaviors of the participants to those of nonparticipants and found that the program is effective in delaying the initiation of sexual intercourse as well as increasing the use of contraception by those who were sexually active (Dixon, Schoonmaker, and Philliper 2000). Research suggests that culturally sensitive information can improve the healthcare message regarding sexual intercourse and the role of contraception, ultimately reducing the teen pregnancy rate for African-Americans.
The research on teen pregnancy, specifically on minority teen pregnancy rates, leads us to hypothesize that the passive representation of minority teachers will decrease minority teen pregnancy rates. Given the importance of cultural sensitivity, we expect that race-match representation will have the greatest impact on African-American pregnancy rates.3 The hypothesis for female teachers is less clear. Although previous research in other policy areas suggests we should expect that female teachers will decrease the pregnancy rate among all students, the nature of this policy area causes us to reconsider. In their research on the impact of healthcare resources and teen birth rates, Walker and Zhu (2010) find that increased monetary input and healthcare human capital significantly decrease the birth rates for white teens, but not for African-American teens. In fact, the benefit gap between African-Americans and whites widens as healthcare human capital increases. Walker and Zhu posit that this finding may be due to a lack of racial representation in state healthcare systems, which employ primarily white women. Additionally, they argue that neutrally designed policies might produce biased policy outcomes for minorities. Consistent with these findings and previous work (Dixon, Schoonmaker, and Philliper, 2000; Tabi 2002), we posit that the passive representation of female teachers will not be related to teen pregnancy rates.
Insights from Interviews with Teachers and Administrators
To broaden and deepen our understanding of the possible causal mechanisms that underpin our hypotheses, we interviewed a convenience sample of 11 high-school teachers and one school district administrator. We asked the teachers and administrator to meet with us to discuss the influence that teachers can have on student behavior both inside and outside the classroom. The teachers represent several high schools, in both majority–majority and majority–minority settings. We were able to speak with three African-American women, four white women, and four white men.4 Our discussions with teachers and administrator offer insights into the influence that teachers have in the decisions (educational and noneducational) of their students.
Evidence of Role Models
Recall that one of the causal mechanisms suggested in the literature is that minority or female bureaucrats serve as roles models for clients who share their demographic characteristics. All of the teachers we spoke with, regardless of race or sex, discussed how they serve as role models for students regarding educational choices. One white male teacher told us that students often ask him where he went to college and how he decided where to go. This was a common theme across all of the teachers we spoke with; however, the discussions with African-American teachers took on a slightly different character. One African-American female teacher talked about how ten years ago, when she started teaching eighth-grade English, many of her African-American students commented she was the first African-American teacher they had a class with. She went on to say how important it is for African-American students to see everybody represented in the school, so that they know they can do anything. She spent time discussing her own experiences and the obstacles she has been able to overcome with her students. Similarly, another African-American female teacher said that numerous African-American students stop by her room to discuss their future plans, and she attributes this to the students feeling like she can understand their path and identify with their circumstances.
We found additional evidence of the importance of a race-match when we asked teachers about serving as role models for noneducational outcomes, specifically in relationship choices. Two white male teachers discussed how they are open about their marriages and role as fathers in hopes of modeling good behavior for their students, especially male students. However, both of these teachers said that although they willingly discuss these issues they get questions regularly from white students but seldom from African-American male or female students. The case was quite different for African-American female teachers. All of the African-American female teachers we spoke with shared example after example of both male and female students asking questions about relationship choices and decisions. For example, one African-American teacher said some African-American female students asked her if she chose to have premarital sex and if she thought it was all right for them to have sex with their boyfriends. Similarly, another African-American female teacher said that African-American male students ask her about how to treat their girlfriends and about parenting. These questions lead to frank discussions about contraception, pregnancy, and risky behaviors. Our discussions convinced us that, although any teacher can serve as a role model, African-American students seek out role models that look like them, particularly with regard to noneducational issues.
Evidence of Shaping Policy and Practice
Another possible way that teachers can influence noneducational outcomes, specifically reproductive health outcomes, of students is by shaping the policies and practices surrounding sex education in the school district they teach in. We find strong support for this causal mechanism in our interviews with teachers and an administrator. The white female administrator we spoke with said that there is definitely a role for teachers in the development of district-level sex education. She stated that “teachers are encouraged to be part of the discussion, especially those from underrepresented groups.” She also said that teachers collaborate with community groups, like the Pregnancy Task Force, to address the district’s concerns about teenage pregnancy. She credited minority teachers’ input in the design of the district’s culturally sensitive sex education program. However, the teachers we spoke with felt that they have a limited voice in formal policy, but can make a significant difference through practice. One African-American female teacher discussed a sex education program that she initiated with some students. She said that several African-American female students approached her to say that they wanted to do something about all of the “pregnant bellies” that popped out around school each spring. The teacher and students developed an after-school program that both African-American male and female students attended. The program involved role-playing and scenarios designed to give the students skills to deal with sexual situations and to avoid risky sexual behaviors. The teacher felt confident that this program was beneficial to all of the students involved. The program caught the attention of the school’s administration, which considered implementing it on a broader scale, but a change in leadership derailed those prospects. The African-American female teacher remains committed to continuing the program with interested students. Another African-American female teacher discussed the pregnancy prevention program she created for the predominantly African-American female sport team she coaches. The program garnered high praise from the athletes who requested that she continue it in the following school year. In addition to any involvement at the district level, we find that African-American teachers are finding important ways to influence the behaviors that lead to pregnancy.
Evidence of Mentoring
An alternative explanation of how the presence of African-American teachers may improve the reproductive health of African-American students is that African-American teachers build relationships with African-American students, which allow them to serve as trusted advisors and mentors for the students. We found overwhelming support for this in our discussions with teachers and administrator. The white teachers and the administrator indicated that they find African-American students are more comfortable and build tighter bonds with African-American teachers. They point to a shared culture, experience, and language as the likely reasons for this, even if this match is only perceived. However, most of them conceded that the power of “looking like” the students should not be underestimated. The African-American female teachers we interviewed each had numerous examples of African-American students seeking advice and guidance on issues related to reproductive health. These teachers stress that African-American students are more comfortable with them because the students assume that African-American teachers come from a similar background and that these teachers will understand them. Due to this, African-American students, most often females but also males, trust these teachers and are comfortable having candid conversations about sex, birth control, STDs, and relationships, which can reduce the risk of teen pregnancy.
White teachers talk about the extra effort they make to build relationships with African-American students, but admit they have not been very successful. For example, one white male teacher said that he tries to be “wide open” with students and available to talk with them; however, race is a huge factor because, “as a white guy,” African-American students do not believe he can understand them. In fact, students would tell him “you did not grow up in my house” or comment “you ain’t my daddy” when he tries to provide advice. He says he is sometimes successful in building a relationship with African-American female students but seldom with males. Another white male teacher said that he has a tougher time building relationships with African-American students than with white or Latina students. Given this, he takes a different approach; he coaches and serves as the graduation coordinator in hopes that if students see him in the classroom, on the field, and in other settings, they, regardless of race, will begin to connect with him. This evidence suggests that a race-match is critical in building mentoring relationships, which can ultimately produce positive outcomes for African-American students.
Evidence of Changing the Behavior of Majority Bureaucrats
Scholars hypothesize that the presence of minority bureaucrats can change the behavior and/or attitudes of the majority bureaucrats who work with them. We find support for this argument in our interviews with teachers. The white female administrator we spoke with said that the presence of African-American teachers places pressure on the school district to be culturally aware and sensitive as they develop policies. In addition, several white teachers, both males and females, identify African-American female teachers in their schools who they seek advice from when handling sensitive situations (ranging from dress code violations to handling a distraught pregnant student). The white teachers also said that they, if do not feel like they can properly handle a situation, refer the African-American students (most often females) to these African-American female teachers. The white teachers added that without these African-American female teachers to “provide backup” they would not feel as comfortable dealing with delicate topics. Overall, these interviews lead us to believe that several of the causal mechanisms highlighted in the literature are at work. In the remainder of the article we empirically test whether the presence of African-American teachers can lower teen pregnancy rates for African-American students.
Data and Method
In the state of Georgia, school district jurisdictions overlap with county boundaries. This allows us to test the influence of school district–level characteristics on county-level outcomes, such as the teenage pregnancy rate. To test the relationship between the presence of female and minority teachers and pregnancy rates of students, we combine data from three sources. We use data from public schools in the state of Georgia for the school years 2002–3, 2003–4, 2004–5, and 2005–6. These data provide an excellent opportunity for examining the influence of representation in the public sector for several reasons. Representation variables are easily constructed at the district level for both teachers and administrators; student data are readily available; and data are available for numerous control variables that help to account for the influence of the school context. The data on teen pregnancy rates are from the Georgia Department of Community Health’s Online Analytical Statistical Information System (OASIS). Data from the National Center for Education Statistics (NCES) are used to supplement the Georgia educational data by adding a measure of district-level resources. Finally, Census data are used to gather unemployment rates to be used in the analysis. The unit of analysis is school districts in the state of Georgia.
The outcome variable we focus on is the pregnancy rate for young women aged 15–19 years in the county each year. Pregnancy rate is measured as the number of pregnancies per 1,000 females ages 15–19. We include pregnancy rates by race for white and African-American females from 2002 to 2005. We are specifically interested in the relationship between representation of teachers at high schools in a district and the teenage pregnancy rate for the county. For this reason, we do not include the pregnancy rate for young women aged 10–14 years. Although some of the pregnancies in this age group would qualify as teenage pregnancies, some would not, and it is not possible to disaggregate the data further by age. Additionally, the majority of teenage pregnancies occur in the 15–19-year-old sample. The state-level pregnancy rate for all races in our sample for 10–14-year-olds is 1.7 (2,156 pregnancies), whereas the state-level pregnancy rate for all races in the 15–19-year-old sample is 70.2 (84,633 pregnancies).5
The national pregnancy rate for young women aged 15–19 years declined from 76.0 in 2002 to 70.6 in 2005 (Ventura et al. 2009). The average pregnancy rates were 86.21 for African-American and 58.70 for white young women aged 15–19 in Georgia from 2002 to 2005 (see table 1). We have data for 143 school districts over a 4-year period. Once missing data6 are accounted for, this results in a final sample size of 506 district-year observations.7
Variable Descriptive Statistics
| Mean | SD | Minimum | Maximum | |
|---|---|---|---|---|
| Black pregnancy rate | 86.21 | 35.89 | 0.00 | 240.70 |
| White pregnancy rate | 58.70 | 21.52 | 0.00 | 129.30 |
| Percent female teachers | 60.97 | 6.91 | 36.84 | 77.78 |
| Percent African-American teachers | 13.40 | 13.18 | 0.00 | 83.64 |
| Percent Latina teachers | 0.73 | 1.07 | 0.00 | 6.67 |
| Percent white teachers | 85.39 | 13.90 | 14.55 | 100.00 |
| Percent African-American students | 36.44 | 22.52 | 0.00 | 98.44 |
| Percent Latina students | 2.65 | 3.12 | 0.00 | 20.35 |
| Percent white students | 59.22 | 21.92 | 1.08 | 99.01 |
| Total current expenditures per student (US$) | 4886.31 | 431.55 | 3795.00 | 6383.00 |
| Student–teacher ratio | 15.82 | 1.76 | 8.08 | 23.54 |
| Percent free and reduced lunch | 44.36 | 17.74 | 3.41 | 100.00 |
| Unemployment rate | 5.21 | 1.13 | 3.30 | 10.00 |
| 0%–9% African-American teachers | 0.53 | 0.50 | 0.00 | 1.00 |
| 10%–19% African-American teachers | 0.23 | 0.42 | 0.00 | 1.00 |
| 20%–29% African-American teachers | 0.12 | 0.32 | 0.00 | 1.00 |
| 30%–39% African-American teachers | 0.06 | 0.24 | 0.00 | 1.00 |
| Over 40% African-American teachers | 0.05 | 0.22 | 0.00 | 1.00 |
| Observations | 506 | |||
| Mean | SD | Minimum | Maximum | |
|---|---|---|---|---|
| Black pregnancy rate | 86.21 | 35.89 | 0.00 | 240.70 |
| White pregnancy rate | 58.70 | 21.52 | 0.00 | 129.30 |
| Percent female teachers | 60.97 | 6.91 | 36.84 | 77.78 |
| Percent African-American teachers | 13.40 | 13.18 | 0.00 | 83.64 |
| Percent Latina teachers | 0.73 | 1.07 | 0.00 | 6.67 |
| Percent white teachers | 85.39 | 13.90 | 14.55 | 100.00 |
| Percent African-American students | 36.44 | 22.52 | 0.00 | 98.44 |
| Percent Latina students | 2.65 | 3.12 | 0.00 | 20.35 |
| Percent white students | 59.22 | 21.92 | 1.08 | 99.01 |
| Total current expenditures per student (US$) | 4886.31 | 431.55 | 3795.00 | 6383.00 |
| Student–teacher ratio | 15.82 | 1.76 | 8.08 | 23.54 |
| Percent free and reduced lunch | 44.36 | 17.74 | 3.41 | 100.00 |
| Unemployment rate | 5.21 | 1.13 | 3.30 | 10.00 |
| 0%–9% African-American teachers | 0.53 | 0.50 | 0.00 | 1.00 |
| 10%–19% African-American teachers | 0.23 | 0.42 | 0.00 | 1.00 |
| 20%–29% African-American teachers | 0.12 | 0.32 | 0.00 | 1.00 |
| 30%–39% African-American teachers | 0.06 | 0.24 | 0.00 | 1.00 |
| Over 40% African-American teachers | 0.05 | 0.22 | 0.00 | 1.00 |
| Observations | 506 | |||
Variable Descriptive Statistics
| Mean | SD | Minimum | Maximum | |
|---|---|---|---|---|
| Black pregnancy rate | 86.21 | 35.89 | 0.00 | 240.70 |
| White pregnancy rate | 58.70 | 21.52 | 0.00 | 129.30 |
| Percent female teachers | 60.97 | 6.91 | 36.84 | 77.78 |
| Percent African-American teachers | 13.40 | 13.18 | 0.00 | 83.64 |
| Percent Latina teachers | 0.73 | 1.07 | 0.00 | 6.67 |
| Percent white teachers | 85.39 | 13.90 | 14.55 | 100.00 |
| Percent African-American students | 36.44 | 22.52 | 0.00 | 98.44 |
| Percent Latina students | 2.65 | 3.12 | 0.00 | 20.35 |
| Percent white students | 59.22 | 21.92 | 1.08 | 99.01 |
| Total current expenditures per student (US$) | 4886.31 | 431.55 | 3795.00 | 6383.00 |
| Student–teacher ratio | 15.82 | 1.76 | 8.08 | 23.54 |
| Percent free and reduced lunch | 44.36 | 17.74 | 3.41 | 100.00 |
| Unemployment rate | 5.21 | 1.13 | 3.30 | 10.00 |
| 0%–9% African-American teachers | 0.53 | 0.50 | 0.00 | 1.00 |
| 10%–19% African-American teachers | 0.23 | 0.42 | 0.00 | 1.00 |
| 20%–29% African-American teachers | 0.12 | 0.32 | 0.00 | 1.00 |
| 30%–39% African-American teachers | 0.06 | 0.24 | 0.00 | 1.00 |
| Over 40% African-American teachers | 0.05 | 0.22 | 0.00 | 1.00 |
| Observations | 506 | |||
| Mean | SD | Minimum | Maximum | |
|---|---|---|---|---|
| Black pregnancy rate | 86.21 | 35.89 | 0.00 | 240.70 |
| White pregnancy rate | 58.70 | 21.52 | 0.00 | 129.30 |
| Percent female teachers | 60.97 | 6.91 | 36.84 | 77.78 |
| Percent African-American teachers | 13.40 | 13.18 | 0.00 | 83.64 |
| Percent Latina teachers | 0.73 | 1.07 | 0.00 | 6.67 |
| Percent white teachers | 85.39 | 13.90 | 14.55 | 100.00 |
| Percent African-American students | 36.44 | 22.52 | 0.00 | 98.44 |
| Percent Latina students | 2.65 | 3.12 | 0.00 | 20.35 |
| Percent white students | 59.22 | 21.92 | 1.08 | 99.01 |
| Total current expenditures per student (US$) | 4886.31 | 431.55 | 3795.00 | 6383.00 |
| Student–teacher ratio | 15.82 | 1.76 | 8.08 | 23.54 |
| Percent free and reduced lunch | 44.36 | 17.74 | 3.41 | 100.00 |
| Unemployment rate | 5.21 | 1.13 | 3.30 | 10.00 |
| 0%–9% African-American teachers | 0.53 | 0.50 | 0.00 | 1.00 |
| 10%–19% African-American teachers | 0.23 | 0.42 | 0.00 | 1.00 |
| 20%–29% African-American teachers | 0.12 | 0.32 | 0.00 | 1.00 |
| 30%–39% African-American teachers | 0.06 | 0.24 | 0.00 | 1.00 |
| Over 40% African-American teachers | 0.05 | 0.22 | 0.00 | 1.00 |
| Observations | 506 | |||
The key variables of interest in this analysis are measures of representation including teacher race/ethnicity and sex. We include the percent African-American teachers in a district, the percent white teachers in a district, and the percent female teachers in a district as measures of representation. We expect that passive representation of minority teachers will decrease the minority teen pregnancy rates. Our expectation for the effect of female teacher passive representation on teenage pregnancy rates (minority and nonminority) is less clear.
In addition to the key independent variables, we include several control variables to account for other factors that may influence teen pregnancy rates. We include measures of unemployment rate to control for socioeconomic differences among districts. We also add a number of controls in an effort to pick up key institutional differences among districts. Because poverty and race are often correlated with negative educational outcomes, we include the percentage of students in the district who received free and reduced lunch as a measure of poverty and the percentage of African-American and Latina students in a district to account for differences in organizational task difficulty faced by the schools. In addition, organizational resources are also correlated with educational outcomes for students. In order to account for this relationship, we control for current instructional expenditures per student in each district as well as student–teacher ratio. Because our data are gathered across 4 years, we include a set of time series dummy variables to control for differences across time.
In some specifications, we choose to control for white pregnancy rate in order to pick up a number of factors that would be difficult (if not impossible) to quantify.8 Including this variable controls for unmeasured factors that affect the African-American teen pregnancy rates. For the two models predicting African-American pregnancy rates, we control for white pregnancy rates. Although this significantly raises the hurdle for achieving statistical significance, we believe that our models would be severely underspecified without controlling for white pregnancy rates in modeling the relationship between coethnic representation for minorities and pregnancy rates. The descriptive statistics for the variables used in our analysis are provided in table 1.
We estimate the effect of race-match representation on the teenage pregnancy rates for African-American and white students in a district. We perform three different specifications for the models predicting the African-American teen pregnancy rate in a district. The first specification predicts the African-American teen pregnancy rate as a function of a vector of representation variables for district d in year t, Xtd, which includes the percentage of teachers in the district who are African-American, Latina, and female. The control variables in this specification include vectors of district-level school and community characteristics for district d in time t, Ytd. The model specification for African-American teen pregnancy rate is:
The second model predicting the African-American pregnancy rate is specified in the same way; however, we add the white teen pregnancy rate in the district as a control. Because theory suggests that a critical mass of minority bureaucrats may be needed to influence outcomes (Meier 1993), in the third and preferred specification, we allow for a nonlinear relationship between the percentage of African-American teachers in a district and African-American teen pregnancy rate by separating the percentage of African-American teachers into ranges.9 We also control for white pregnancy rate in this final specification.
The next set of results focuses on the white teen pregnancy rate as the outcome. We include this regression in the analysis in order to test whether representation mattered for nonminority students and whether or not any tradeoffs might occur between minority teacher representation and outcomes for nonminority students. The specification is similar to the previously outlined equation; however, the representation variables include the percentage of teachers in the district who are white, African-American, and female. The control variables are the same as the African-American pregnancy rate regressions.
Each of the previously discussed specifications is estimated using ordinary least squares (OLS) regression.10 Because it is likely that the error terms within a district are correlated, we cluster the standard errors at the district level.11
Results and Discussion
The multivariate analysis of African-American teen pregnancy rates is presented in table 2. As discussed above, existing work would lead us to expect that this relationship between the representation of African-American teachers and African-American teen pregnancy rates would be negative and significant. Our findings support this hypothesis. In the first model, we find that the linear term for African-American teachers is negative and statistically significant. Increased African-American teacher representation is correlated with a decrease in African-American teen pregnancy rates. Although this relationship is not large when calculated at the mean values of African-American teacher representation and African-American pregnancy rates, it is still meaningful. For the linear model, at the mean values, we find that a 1% increase in the representation of African-American teachers is associated with a 0.55 decrease in African-American teen pregnancies per 1,000 African-American young women aged 15–19 years.12 In other words, a 10% increase in the representation of African-American teachers would result in approximately six fewer African-American teen pregnancies in a district. However, we will show that results are strongly nonlinear and we observe much stronger effects for districts further away from the means. The passive representation of female teachers is not related to the African-American teen pregnancy rate. Again, this may be due to the fact that 82% of African-American teachers in Georgia are female (Nweke et al. 2004). In the second model, we add a control for the white pregnancy rate and find that it is positively related to pregnancy rates for African–American teens. However, the linear term for African-American teachers remains negative and significant. Model 3 offers our preferred specification and examines the possible nonlinear relationship between the presence of African-American teachers and the pregnancy rate for African-American teens.
The Relationship between Teacher Representation and African-American Pregnancy Rates (OLS regression results)
| Model 1 | Model 2 | Model 3 | |
|---|---|---|---|
| Percent female teachers | 0.48 (1.61) | 0.44 (1.56) | 0.36 (1.35) |
| Percent African-American teachers | −0.55** (−2.26) | −0.61** (−2.58) | |
| 10%–19% African-American teachers | −4.28 (−0.91) | ||
| 20%–29% African-American teachers | −18.8** (−2.49) | ||
| 30%–39% African-American teachers | −28.6*** (−3.12) | ||
| Over 40% African-American teachers | −34.7*** (−3.19) | ||
| Percent Latina teachers | −1.50 (−0.82) | −1.40 (−0.83) | −1.47 (−0.94) |
| Percent African-American students | 0.69** (2.58) | 0.86*** (3.34) | 0.98*** (3.88) |
| Percent Latina students | 1.26* (1.88) | 0.92 (1.33) | 1.05 (1.52) |
| Logged total current expenditures per student | −37.3 (−1.33) | −30.0 (−1.06) | −26.9 (−0.96) |
| Student–teacher tatio | 0.71 (0.53) | 0.57 (0.44) | 0.59 (0.46) |
| Percent free and reduced lunch | 0.17 (0.74) | −0.072 (−0.31) | −0.12 (−0.51) |
| Unemployment rate | 5.43** (2.15) | 4.96** (2.01) | 4.67* (1.87) |
| White pregnancy rate | 0.30*** (2.78) | 0.32*** (3.04) | |
| Observations | 506 | 506 | 506 |
| Model 1 | Model 2 | Model 3 | |
|---|---|---|---|
| Percent female teachers | 0.48 (1.61) | 0.44 (1.56) | 0.36 (1.35) |
| Percent African-American teachers | −0.55** (−2.26) | −0.61** (−2.58) | |
| 10%–19% African-American teachers | −4.28 (−0.91) | ||
| 20%–29% African-American teachers | −18.8** (−2.49) | ||
| 30%–39% African-American teachers | −28.6*** (−3.12) | ||
| Over 40% African-American teachers | −34.7*** (−3.19) | ||
| Percent Latina teachers | −1.50 (−0.82) | −1.40 (−0.83) | −1.47 (−0.94) |
| Percent African-American students | 0.69** (2.58) | 0.86*** (3.34) | 0.98*** (3.88) |
| Percent Latina students | 1.26* (1.88) | 0.92 (1.33) | 1.05 (1.52) |
| Logged total current expenditures per student | −37.3 (−1.33) | −30.0 (−1.06) | −26.9 (−0.96) |
| Student–teacher tatio | 0.71 (0.53) | 0.57 (0.44) | 0.59 (0.46) |
| Percent free and reduced lunch | 0.17 (0.74) | −0.072 (−0.31) | −0.12 (−0.51) |
| Unemployment rate | 5.43** (2.15) | 4.96** (2.01) | 4.67* (1.87) |
| White pregnancy rate | 0.30*** (2.78) | 0.32*** (3.04) | |
| Observations | 506 | 506 | 506 |
Note: Time trend included but not shown.
* p < .1, ** p < .05, *** p < .01.
The Relationship between Teacher Representation and African-American Pregnancy Rates (OLS regression results)
| Model 1 | Model 2 | Model 3 | |
|---|---|---|---|
| Percent female teachers | 0.48 (1.61) | 0.44 (1.56) | 0.36 (1.35) |
| Percent African-American teachers | −0.55** (−2.26) | −0.61** (−2.58) | |
| 10%–19% African-American teachers | −4.28 (−0.91) | ||
| 20%–29% African-American teachers | −18.8** (−2.49) | ||
| 30%–39% African-American teachers | −28.6*** (−3.12) | ||
| Over 40% African-American teachers | −34.7*** (−3.19) | ||
| Percent Latina teachers | −1.50 (−0.82) | −1.40 (−0.83) | −1.47 (−0.94) |
| Percent African-American students | 0.69** (2.58) | 0.86*** (3.34) | 0.98*** (3.88) |
| Percent Latina students | 1.26* (1.88) | 0.92 (1.33) | 1.05 (1.52) |
| Logged total current expenditures per student | −37.3 (−1.33) | −30.0 (−1.06) | −26.9 (−0.96) |
| Student–teacher tatio | 0.71 (0.53) | 0.57 (0.44) | 0.59 (0.46) |
| Percent free and reduced lunch | 0.17 (0.74) | −0.072 (−0.31) | −0.12 (−0.51) |
| Unemployment rate | 5.43** (2.15) | 4.96** (2.01) | 4.67* (1.87) |
| White pregnancy rate | 0.30*** (2.78) | 0.32*** (3.04) | |
| Observations | 506 | 506 | 506 |
| Model 1 | Model 2 | Model 3 | |
|---|---|---|---|
| Percent female teachers | 0.48 (1.61) | 0.44 (1.56) | 0.36 (1.35) |
| Percent African-American teachers | −0.55** (−2.26) | −0.61** (−2.58) | |
| 10%–19% African-American teachers | −4.28 (−0.91) | ||
| 20%–29% African-American teachers | −18.8** (−2.49) | ||
| 30%–39% African-American teachers | −28.6*** (−3.12) | ||
| Over 40% African-American teachers | −34.7*** (−3.19) | ||
| Percent Latina teachers | −1.50 (−0.82) | −1.40 (−0.83) | −1.47 (−0.94) |
| Percent African-American students | 0.69** (2.58) | 0.86*** (3.34) | 0.98*** (3.88) |
| Percent Latina students | 1.26* (1.88) | 0.92 (1.33) | 1.05 (1.52) |
| Logged total current expenditures per student | −37.3 (−1.33) | −30.0 (−1.06) | −26.9 (−0.96) |
| Student–teacher tatio | 0.71 (0.53) | 0.57 (0.44) | 0.59 (0.46) |
| Percent free and reduced lunch | 0.17 (0.74) | −0.072 (−0.31) | −0.12 (−0.51) |
| Unemployment rate | 5.43** (2.15) | 4.96** (2.01) | 4.67* (1.87) |
| White pregnancy rate | 0.30*** (2.78) | 0.32*** (3.04) | |
| Observations | 506 | 506 | 506 |
Note: Time trend included but not shown.
* p < .1, ** p < .05, *** p < .01.
Given the theory of critical mass, we expect that the representation of African-American teachers will be related in a curvilinear fashion to African-American teen pregnancy rates. We model this relationship using representation categories because it is flexible and allows for different effects across differing levels of representation. Initially, teacher representation would not be related to teen pregnancy rates, but would become more favorable as the representation of teachers reached a critical mass. The comparison group for the representation categories is districts with less than 10% African-American teachers. Being in a district with 10%–19% African-American teachers does not decrease the pregnancy rate relative to a district with less than 10% African-American teachers. However, once a district reaches a threshold level of 20%–29% African-American teachers, we observe a significant decrease in African-American teen pregnancy rates (18.8 fewer pregnancies per 1,000). Interestingly, this relationship is monotonic in nature. The number of pregnancies continues to drop as representation increases. When there are few African-American teachers in a school, we observe no effect on African-American teen pregnancies. However, this relationship eventually becomes negative as the representation of African-American teachers increases. Other variables in our model are also related to the pregnancy rate among African-American teens. In the preferred model, increased African-American student population, unemployment, and white teen pregnancy rate are all associated with higher African-American pregnancy rates.
Finally, the models predicting the white teen pregnancy rate are presented in table 3. The representation variables for all groups (female, white, and African-American) of teachers fail to reach statistical significance. One notable aspect of this finding is that increasing African-American teacher representation does not create a tradeoff in terms of lowering the African-American teen pregnancy rate, but increasing this rate for white teens. The variables that are positively and significantly related to white teen pregnancy rates are the percentage of students who are Latina and the percentage of students who are receiving free and reduced lunch. On the other hand, an increase in the percentage of students who are African-American is associated with a lower white pregnancy rate.
The Relationship between Teacher Representation and White Pregnancy Rates (OLS regression results)
| Model 1 | |
|---|---|
| Percent female teachers | 0.12 (0.58) |
| Percent white teachers | 0.34 (0.34) |
| Percent African-American teachers | 0.52 (0.47) |
| Percent African-American students | −0.56*** (−3.47) |
| Percent Latina students | 1.16** (2.16) |
| Logged total current expenditures per student | −25.0 (−1.40) |
| Student–teacher ratio | 0.48 (0.72) |
| Percent free and eeduced lunch | 0.81*** (6.22) |
| Unemployment rate | 1.57 (1.03) |
| Observations | 506 |
| Model 1 | |
|---|---|
| Percent female teachers | 0.12 (0.58) |
| Percent white teachers | 0.34 (0.34) |
| Percent African-American teachers | 0.52 (0.47) |
| Percent African-American students | −0.56*** (−3.47) |
| Percent Latina students | 1.16** (2.16) |
| Logged total current expenditures per student | −25.0 (−1.40) |
| Student–teacher ratio | 0.48 (0.72) |
| Percent free and eeduced lunch | 0.81*** (6.22) |
| Unemployment rate | 1.57 (1.03) |
| Observations | 506 |
Note: Time trend included but not shown.
* p < .1, ** p < .05, *** p < .01.
The Relationship between Teacher Representation and White Pregnancy Rates (OLS regression results)
| Model 1 | |
|---|---|
| Percent female teachers | 0.12 (0.58) |
| Percent white teachers | 0.34 (0.34) |
| Percent African-American teachers | 0.52 (0.47) |
| Percent African-American students | −0.56*** (−3.47) |
| Percent Latina students | 1.16** (2.16) |
| Logged total current expenditures per student | −25.0 (−1.40) |
| Student–teacher ratio | 0.48 (0.72) |
| Percent free and eeduced lunch | 0.81*** (6.22) |
| Unemployment rate | 1.57 (1.03) |
| Observations | 506 |
| Model 1 | |
|---|---|
| Percent female teachers | 0.12 (0.58) |
| Percent white teachers | 0.34 (0.34) |
| Percent African-American teachers | 0.52 (0.47) |
| Percent African-American students | −0.56*** (−3.47) |
| Percent Latina students | 1.16** (2.16) |
| Logged total current expenditures per student | −25.0 (−1.40) |
| Student–teacher ratio | 0.48 (0.72) |
| Percent free and eeduced lunch | 0.81*** (6.22) |
| Unemployment rate | 1.57 (1.03) |
| Observations | 506 |
Note: Time trend included but not shown.
* p < .1, ** p < .05, *** p < .01.
Conclusion
This article examines the relationship between the passive representation of minority and female bureaucrats and pro-social outcomes beyond the immediate scope of the organization. Specifically, we explore how the bureaucratic representation of minority and female teachers can improve reproductive health outcomes for female students. We find support for our hypothesis that increasing the representation of African-American teachers significantly decreases the African-American teen pregnancy rates. Further analysis demonstrates that the relationship between the representation of African-American teachers and African-Americans teen pregnancy rates is curvilinear. For the decrease in pregnancy rates to occur, African-American teachers need to reach a critical mass of at least 20%. This relationship does not hold for white teachers—the percentage of white teachers does not appear to be related to the pregnancy rate for white teens.
Given the seriousness of the teen pregnancy problem in this country, it is important to identify strategies to combat it. This is especially true for the African-American population, which has the highest rate of teen pregnancy among all groups. Increasing the representation of African-American teachers, both male and female, appears to have a positive impact on African-American students. We should note that this influence is not limited to the interaction between African-American teachers and female students. There is evidence that the presence of African-American male teachers has a positive effect on the sexual behavior of African-American male students (Rezai-Rashti and Martino 2010). Given this, schools should consider how their hiring decisions might be related to the reproductive health of their students.
Although we do not know with certainty the causal mechanism underlying these findings, we can identify several possible pathways supported by our qualitative analysis. Our interviews indicate that African-American teachers serve as mentors, take a special interest in the behaviors and decisions of African-American students, and actively try to have an influence on them. These teachers talk candidly with students about sexual behavior and offer valuable advice. The presence of African-American teachers influences the behavior of teachers and administrators from other demographic groups by making them more sensitive to the issue of teen pregnancy in the African-American community. Similarly, the priorities and policies of the school are influenced by the presence of African-American teachers. Finally, African-American teachers serve as role models for African-American students, which can encourage students to make healthy decisions regarding sexual activity.
One caveat of this study is that it is difficult to know how generalizable the findings are. These variables, models, and findings are context-specific, and it is not clear whether the results could be generalized to other policy settings or bureaucrats other than teachers. However, following Keiser et al. (2002), we could expect representation to affect outcomes beyond the agency when the issue is salient, the bureaucrat has discretion, and the institutional setting promotes the provision of active representation. It is likely that the representation of other types of bureaucrats (police officers, counselors, social workers) may also produce pro-social outcomes. It is also probable that the representation of minority teachers is related to improving other pro-social outcomes (e.g., crime rates). Nevertheless, we can also imagine situations where this relationship is unlikely to exist. A second limitation is that we use data from public schools in only one state. Although it is a fairly large and diverse state, it would still be ideal to have data from all states to provide the most generalizable set of findings. Future research is necessary to identify the conditions under which we can expect the representation of minority or female bureaucrats to improve outcomes that extend beyond the organization.
References
We would concede that teen pregnancy is definitely part of the education production function and should be a priority of schools. However, our interviews with teachers confirm that administrators do not recognize and/or respond to the link between educational outcomes and reproductive health. In a high-stakes testing environment, schools treat pregnancy and its prevention as a secondary or tertiary concern.
Interestingly, Gay’s (2001) findings suggest that the benefits of symbolic representation are limited to minorities. However, the findings of Theilemann and Stewart (1996) challenge this notion.
Although we expect the same relationship to exist between Latina pregnancy rates and the presence of Latina teachers, the limited number of Latina teachers working in Georgia schools makes it impossible to analyze this relationship with these data.
Regrettably we were unable to interview any African-American male teachers. In the state of Georgia, 82% of the African-American school teachers are female (Nweke et al. 2004). Given this, our race variable is really a measure of African-American female teachers.
By including 18- and 19-year-old females in the pregnancy rate measure, we may capture pregnancies to nonstudents. This could also capture pregnancies to high school dropouts. Any bias produced by these types of measurement error would likely drive our coefficients on the representation variables toward zero since these young women would not be subject to the treatment (representation of teachers).
Our sample size drops from 608 to 506 due to missing data on several variables—teacher representation, student–teacher ratio, pregnancy rates, student demographics, and free and reduced lunch. These variables appear to be missing at random, and the districts dropped from our data are not significantly different from those included in our analysis.
We also performed a sensitivity analysis by limiting the sample to school districts with only one high school, allowing us to test the relationship between teacher representation and pregnancy rates at the school level. This provided a direct match between teacher representation variables and the outcomes of interest. In this sub-sample, we have a sample size of 317 school-year observations. We performed the same analysis on a sub-sample of districts with a single high school, and the results are consistent with the finding for the full sample.
We recognize that including the white pregnancy rate as a predictor of the African-American teen pregnancy rate could introduce endogeneity bias into our parameter estimates. If the correlation between the white pregnancy rate and the error term is not equal to zero and the percentage of African-American teachers and the white pregnancy rate are correlated, this would bias the parameter estimate for the percentage of African-American teachers. We did not have data for an instrumental variables regression to address this issue, but we assessed the empirical nature of the problem by running a version of the model that excludes the white pregnancy rate (table 2: models 1 and 2). We found that the magnitudes of the coefficients for the percentage of African-American teachers were stable across models and that the significance levels did not change. The percent change in the coefficients was less than 0.1. This suggests that including the white pregnancy rate as a control variable is not a problem empirically.
One method used to identify a tipping point for a particular variable is to estimate the model with the variable of interest (the percentage of African-American teachers, in our case) squared and take the first derivative of the linear and squared terms. We used this method in earlier drafts and identified that the estimated critical mass of African-American teachers needed to lower the African-American teen pregnancy rate was 17.6%. However, we chose to test this link across several different levels of African-American teacher representation to allow us to fully map this relationship. In addition, this method allowed us to capture the monotonic nature of the nonlinear relationship.
To allow for the fact that the errors in the African-American teen pregnancy rate and white pregnancy rate equations may be correlated, we ran a seemingly unrelated regression, which produced identical results; so for the sake of simplicity we report the results from the OLS regression.
Although we technically have panel data, we did not have sufficient variation within districts to support fixed-effects estimation. Over only a 4-year timeframe, the variables were relatively stable; so there was not enough variation to identify the fixed-effects models. We accounted for correlation in the error term within counties by clustering the standard errors by district.
When calculated as elasticity, we find that a 1% increase in African-American teacher representation is associated with a 0.1% decrease in African-American teen pregnancy rates. To put this into context, Sen (2003) found that increasing alcohol taxes was associated with a decrease in abortion rates and estimated an elasticity of around −0.10, which is comparable to what we find. Levine (2001) estimated the effect of a change in the employment-to-population ratio and found that a 1 percentage point increase in this ratio was associated with a 0.2 percentage point decrease in pregnancy risk for adolescents. Levine’s results are also in line with what we find. Although these studies are not directly comparable to ours, we believe they suggest that increasing minority teacher representation has comparable effect on teen pregnancy.
