The Relationship of School Characteristics and Bullying Between Career Academy and Comprehensive High School Students

Abstract

Violence—including issues of bullying—in urban schools across the United States is a persistent phenomenon, and continues to be of utmost concern for school administrators, community leaders, students, and their families. Using propensity score matching, we examined the relationship between school type—a magnet career academy and a traditional, comprehensive school—and students’ experiences with bullying—observing and being a victim. We found that academy students—compared to comprehensive school students—both experienced and observed significantly lower levels of bullying. We believe it is likely that students at the magnet career academy benefited from a combination of factors, including their abilities to choose to participate in the academy, having peers with similar career interests, and the small size of the academy.

Keywords:

• Bullying
• career academy
• magnet school
• school violence

Violence—including issues of bullying—in urban schools across the United States is a persistent phenomenon, and continues to be of utmost concern for school administrators, community leaders, students, and their families. Researchers have demonstrated that youth residing in low-income urban neighborhoods experience and are exposed to frequent incidents of violence—whether as witnesses or victims (Buka, Stichick, Birdthistle, & Earls, 2001; Duncan, 1996; Mazza & Reynolds, 1999; Okundaye, 2004; Pastore, Fisher, & Friedman, 1996; Paxton, Robinson, Shah, & Schoeny, 2004; Self-Brown, LeBlanc, & Kelley, 2004). Definitions of bullying have varied across research studies (Greif & Furlong, 2006). Olweus definition of bullying is one of the most popular. The researcher defined bullying as peer aggression comprising three tenets: (a) repeated acts over a period of time; (b) with the intent to harm the victim; and (c) having a power imbalance between those who are bullied and their aggressors. Other researchers have added to this definition the following elements: (a) bullying is not provoked by the victim and (b) bullying occurs when other peers are there to witness (Griffin & Gross, 2004). We define bullying as a situation where an aggressive student—or group of students—continually demeans a peer using tactics that include bodily attacks, verbal teasing, and/or social exclusion (Mehta, Cornell, Fan, & Gregory, 2013). As such, bullying in schools contributes to student disengagement, decreased motivation, lower performance, and students’ attempts at avoiding school altogether (Buhs, Ladd, & Herald, 2006; Esbensen & Carson, 2009; Juvonen, Nishina, & Graham, 2000; Schwartz, Gorman, Nakamoto, & Toblin, 2005; Smith, Talamelli, Cowie, Naylor, & Chauhan, 2004). Further, researchers have demonstrated that teasing and bullying in schools are predictors of higher dropout rates (Cornell, Gregory, Huang, & Fan, 2013).

Chronic exposure to various types of violence for urban youth is related to the development of traumatic stress, depression, as well as other serious mental and physical health issues (Lynch, 2003; Margolin & Gordis, 2000; Margolin & Vickerman, 2007; Rossman, Hughes, & Rosenberg, 2000). Chronic exposure to violence also interferes with the healthy psychosocial and neurobiological development of youth (Beers & De Bellis, 2002; Evans & Kim, 2007; Kinniburg, Blaustein, & Spinazzola, 2005; van der Kolk, 2006). Consequently, chronic exposure to violence in adolescence increases their susceptibility to becoming aggressive and violent themselves (Moretti, Obsuth, Odgers, & Reebye, 2006; Pelcovitz, Kaplan, DeRosa, Mandel, & Salzinger, 2000; Schwartz & Proctor, 2000).

In this context, the type of school structure and environment both play major roles in the likelihood of students experiencing and being exposed to school violence. Students who are in urban schools with high percentages of low-income students are more likely to experience school violence (Lleras, 2008). Further, researchers have also studied the impact of school size on the occurrence of bullying, but the relationship between school size and higher incidences of bullying is inconclusive (Cornell et al., 2013; Klein & Cornell, 2010; Lleras, 2008). In general, we know little about how school violence—particularly bullying—differs across school contexts (Wei, Williams, Chen, & Chang, 2010). That is, we have yet to fully understand why violence is higher in some schools compared to others.

In this regard, one popular high school reform initiative in schools across the nation is the career academy. Career academies are small learning communities that features a college-preparatory curriculum with an embedded career theme. Career academies require active partnerships with employers and postsecondary representatives. Career academies reside in all different school configurations, including area/regional career centers, comprehensive schools, and magnet schools. Career academies are designed to increase student engagement and achievement while developing skills necessary to pursue further education or work (Orr et al., 2004). A key condition of career academies is the integration of academic and technical content to increase rigor while building relevancy to students’ career interests (Fletcher et al., 2012; Kemple & Snipes, 2000). There are an estimated 8,000 high schools in the nation offering career academies (MDRC., 2020). Researchers examining students’ experiences in career academies have consistently reported beliefs of being safe within the context of a small community, but quantitative analyses of related behaviors are scant (Fletcher et al., 2019; Kuo, 2010).

Albeit the growing popularity of the career academy concept, the quality of implementation has varied greatly and there have been efforts to inform implementation with the development of standards of practice by school networks such as NAF (formerly known as the National Academy Foundation) (Stern et al., 2010). NAF seeks to promote college and career readiness within the context of occupational themes and postsecondary preparation through customized support to help academies improve and grow (NAF, 2014). NAF evaluates academies on their levels of implementation, using the following hierarchy from highest to lowest: distinguished, model, and certified. NAF continuously evaluates their high school academies to certify their level of implementation based on standards of practice. The academy identified in this research rated as a distinguished academy (the highest level of implementation to the NAF standards of practice) over the past eight years. Also, the academy in this study is a magnet school and focuses on information technology (IT) as a career theme.

The purpose of this study was to examine the relationship between school type—a magnet career academy and a traditional, comprehensive school—and students’ experience with bullying—observing and being a victim. Our research questions included the following:

1. Are there significant differences in academy students’ observing bullying compared to students at a traditional, comprehensive school; and

2. Are there significant differences in academy students’ being a victim of bullying compared to students at a traditional, comprehensive school?

The career academy in this study is part of a large urban school district where students can choose to enroll in traditional, comprehensive high schools with over 2000 students, or magnet schools featuring different structures. With an annual enrollment of about 650 students, the selected academy is one of five magnet school programs in the district, and the only one implemented as a stand-alone school. All other magnet programs operate within traditional high schools.

The impact of school characteristics on students’ experiences and achievement

Research on school size

The school setting and characteristics play a major role in the likelihood of students experiencing and being exposed to school violence. School size is one of the issues that researchers have examined in the literature on bullying. While some researchers have found that school size contributes to the chances of students experiencing violence in school, other researchers’ findings differ. For example, researchers have reported that larger schools contribute to more bullying (Cornell et al., 2013; Ferris & West, 2004; Lleras, 2008; Walker & Gresham, 1997). Related research has revealed the challenges of building positive and supportive cultures in large, comprehensive, urban high schools—particularly those that serve low-income and ethnically and racially diverse youth (Letgers, Balfanz, & McPartland, 2002; Murphy, 2010). To that end, ethnically and racially diverse and low-income students, on average, attend schools located in high criminal activity neighborhoods (Fryer & Levitt, 2004; Land & Legters, 2002; Lee & Burkam, 2002). The problematic conditions of the neighborhoods tend to permeate the culture within schools—creating an unsafe and non-conducive environment for student achievement. As such, the difficulty of establishing a positive culture in comprehensive high schools typically stems from their relatively large student populations and fixed departmental silos. To wit, Letgers et al. (2002) described the context of many large, traditional, comprehensive high schools across the country as follows:

Large size, rigid bureaucratic structures, uninspired teachers, fragmented and irrelevant curriculum, and highly differentiated and unequal learning opportunities have been cited as primary sources of student apathy, alienation, and lack of preparation for college or career. These problems are magnified in high poverty urban high schools that suffer from chronic poor attendance, low achievement, and high dropout rates. (p. 3)

Research on school size has pointed to an ideal range of 600-900 students in a school setting (Lee & Smith, 1997). Hence, one major recommendation that addresses the issue of large schools is the idea of establishing small learning communities. The term “small learning communities” denotes a variety of school structures and configurations—including schools within a school, and magnet programs that are wall-to-wall (where all students participate in a given career theme) (Kuo, 2010). Researchers have found that students in small learning communities experience an increased sense of personalization and belonging, and lower levels of school vandalism (Page, Layzer, Schimmenti, Bernstein, & Horst, 2002). Based on related evidence, Kuo (2010) recommended that: “policymakers and practitioners should continue to find opportunities to reduce the size of large high schools and increase the sense of personalization, belonging, and safety among students, teachers, and staff.” (p. 395)

While some researchers have demonstrated benefits of smaller school sizes, the evidence is mixed. For example, Klein and Cornell (2010) found that while students and teachers in larger schools perceived more bullying in their schools, the actual rates of bullying in these larger schools were lower. Mehta et al. (2013) explained Klein and Cornell (2010) findings stating, “It is likely that students in larger schools were exposed to more bullying incidents simply because there were more students to observe, generating an illusory perception that larger schools were less safe than smaller schools” (p. 46). Further, some researchers have posited that larger schools tend to be more effective and efficient (O’Moore, Kirkham, & Smith, 1997), while others found no relationship between school size and bullying (Khoury-Kassabri, Benbenishty, Astor, & Zeira, 2004; Olweus, 1995).

Research on magnet schools

Besides school size, a limited number of researchers have examined magnet schools as a school factor. Magnet schools were created to utilize incentives instead of coercion to promote racial desegregation as well as to foster individual choice, student diversity, and high quality and innovative educational programs (Frankenberg & Siegel-Hawley, 2008). Researchers have demonstrated that magnet schools are associated with higher rates of student achievement compared to traditional public high schools, private, or religious schools (Ballou, Goldring, & Liu, 2006; Bifulco, Cobb, & Bell, 2008; Gamoran, 1996; Poppell & Hague, 2001). In addition, research has pointed to the need to establish whole school magnets as opposed to school-within-a-school programs –which have been known to racially separate students within the same school (Frankenberg & Siegel-Hawley, 2008). Nonetheless, researchers have not focused on how magnet schools fare as it relates to school violence.

Career themed curricula

In schools, it has been extensively documented that students often find teaching and learning as void of meaning and are prone to question the relevance of instructional tasks (Castellano, Sundell, Overman, & Aliaga, 2012; Hernández-Gantes & Brendefur, 2003). To address this disconnect in teaching and learning, career academies emphasize learning in specific occupational contexts to enhance the relevance of student experience. The premise is that the authenticity of occupational contexts (e.g., IT) provides for opportunities to make learning more meaningful for students (Newmann & Wehlage, 1995; Stipanovic, Lewis, & Stringfield, 2012). Further, researchers have supported the premises of contextual teaching and learning (Newmann, King, & Carmichael, 2007). Researchers have found that students participating in programs emphasizing contextual and authentic teaching and learning showed enhanced student engagement and higher achievement (Newmann et al., 2007; Newmann & Wehlage, 1995). For example, career academy students in magnet programs have the opportunity to select their school based on their interests. Further, these students tend to form a community of learners in school that share their interests. These factors contribute to a positive school experience for students and a sense of belonging (Fletcher & Cox, 2012; Fletcher et al., 2019).

The student experience in career academies

Prior research has revealed that the career academy model has the potential of increasing positive outcomes for students in terms of school attendance, academic course-taking, graduation rates, and dropout rates (Kemple & Snipes, 2000; Newmann et al., 2007; Stipanovic et al., 2012). Recently, researchers have begun to examine the experiences of career academy students to learn more about what contributes to positive student outcomes. For instance—using a case study approach—Fletcher and Cox (2012) studied the experiences and challenges faced by 15 African American academy and non-academy participants. Results emphasized that academy students joined their career academies because they believed that participation would assist them in their preparation for college. Thus, academy students viewed their participation as having value and meaning. They perceived the academy to be beneficial as it created a sense of community within their school, and they viewed their peers and teachers as an extension of their families. On the other hand, non-academy participants chose not to join an academy because they thought it would take time away from engaging in other schooling activities (e.g., co-curricular and extracurricular activities) of interest. Fletcher and Cox also found that academy students were unaware of how their academy courses related to their academic courses.

Fletcher et al. (2019) used a case study approach including 18 classroom observations and 77 interviews to document the school culture of a distinguished NAF wall-to-wall magnet academy of IT. Results pointed to specific key features of the academy model that contributed to the school’s positive school culture. Those features included the open enrollment admissions policy, the nature of a small school, and the shared interest in information technology with peers. Fletcher et al. further found that various constituents of the academy (e.g., district and school administrators, school board members, teachers, school counselors, parents, and community partners) perceived the academy as a safe space for youth and contributed to a sense of equity and inclusion of diverse students.

In addition, Fletcher et al. (2020) compared the student engagement of career academy students to those at a large, traditional comprehensive high school. One of their key findings was that academy students were significantly more likely to be emotionally engaged. Fletcher et al. (2020) defined emotional engagement as students’ perceptions and dispositions toward their school as well as them having a sense of belonging and identification with their school (Appleton, Christenson, & Furlong, 2008; Goodenow, 1993). In addition, emotional engagement included students’ sense of safety at school and their perceived abilities of being able to be themselves within their schools. Mehta et al. (2013) found that bullying was associated with significantly lower levels of student engagement. Mehta et al. (2013) stated that “when students perceive that bullying and teasing is widespread in their school, they feel less safe and become less engaged in their school experience.” (p. 50)

Drawing from this frame of reference, we know that high school career academies have several features that lends itself to a positive school culture and climate (Fletcher et al., 2019). Researchers have demonstrated that career academy students have significantly higher levels of emotional engagement (Fletcher et al., 2020) and that bullying is negatively associated with student engagement in school (Mehta et al., 2013). Given the reports of higher levels of student engagement for academy students and that bullying is negatively related to student engagement, we hypothesized that participation in a career academy would also lead to a lower likelihood of students experiencing and observing school violence—bullying—in their school. Thus, this study is important in our understanding of whether the career academy could be a likely intervention in promoting lower levels of students experiencing and observing bullying in urban high schools. This study contributes to the literature on bullying as it demonstrates the role of career academy’s in reducing the chances of students’ victimization and observations of bullying in school.

Methods

Research design

We used a correlational research design for the implementation of this study, and we collected data using a survey of high school student engagement (Yazzie-Mintz, 2007). In the analysis of our data, we used descriptive statistics to describe the demographic characteristics of our respondents. Next, propensity score matching allowed us to match academy students with students from the comprehensive high school based on key characteristics. Using these matched groups, we computed independent samples t-tests for the mean differences and calculated Cohen’s d effect sizes for the two items on our instrument about bullying. Finally, we used logistic regression modeling to examine the relationship between participation in a high school career academy and their likelihood of experiencing and/or observing bullying.

Data collection procedures

Upon receiving institutional review board (IRB) approval for the study, we identified a school coordinator (e.g., principal, career specialist) to assist with ensuring students understood the study, obtained consent forms, and survey administration. To be eligible to participate in the study, students who were under 18 years of age were required to return signed parental permission and assent forms. For those over the age of 18, we required those students to sign a consent form to participate. Eligible students received an electronic survey link hosted by Qualtrics in April (late spring). The survey duration was approximately 10 to 15 minutes, and students who completed the survey received a $25 electronic Amazon gift card for their time.

Sampling procedure

For comparative purposes, we used purposive sampling to select a distinguished NAF Academy of IT and a large comprehensive school with similar student demographics and achievement—both housed within the same district (Ary, Jacobs, Razavieh, & Sorensen, 2006). The two high schools were in the Southeastern part of the United States. The sites were selected based on the following criteria: (a) NAF Academy of IT representing the highest levels of implementation—distinguished—according to NAF’s annual assessment criteria; (b) a high participation rate of ethnically and racially diverse students as well as a large percentage of students that qualified for free and/or reduced lunch in both schools; and (c) both schools situated within an urban setting. There were 1,162 respondents representing 317 career academy and 845 comprehensive students. Participants were in 9th through 12th grades.

School context

We describe the institutional characteristics, performance outcomes, and student demographics of each school that participated in this study below. We used pseudonyms throughout the manuscript in replacement of school names.

Cascade academy

Cascade Academy of IT is a distinguished—highest level of fidelity/implementation of NAF standards –wall-to-wall NAF academy located in an urban area within the Southeastern region of the United States. Cascade is a magnet school; therefore, students apply to gain admission into it. Because the school has more applicants than seats—the school administrators wanted to keep the school around 650 students—they rely on a lottery system for admission. As of the 2018-2019 data cycle, there were 685 students in the academy with a reported 75.6% to 24.4% male to female split. The gender disparity at Cascade Academy is consistent within the field of IT. White students represented the largest group in the academy at 57%, while Latinx comprising 23%, African American/Black 11%, Asian 7%, and students identified as Other/Multi-Racial at 2%. Cascade Academy of IT reported that 36% of the student body qualified for free and/or reduced lunch prices. For the class of 2017-2018, Cascade reported a 100% graduation rate with 92% of seniors attending a post-secondary institution.

Sampson high school

Sampson High—a traditional comprehensive school—is located in an urban area within the Southeastern region of the United States. Sampson is located in the same school district as Cascade. The school was comprised of approximately 3,380 students. The ethnic and racial backgrounds of students at Sampson High School were as follows: 38% White, 28% African American/Black, 20% Latinx, 11% Asian, and 3% Multiracial. Fifty-seven percent of students qualified for free and/or reduced lunch. The graduation rate for the 2017 to 2018 academic year was 89%.

In terms of student discipline issues in the 2017 to 2018 academic year, Cascade had 20% of students with at least one referral (n = 129 unique students) compared to 30% (n = 957 unique students) at Sampson High School. This school was selected as a comparison to Cascade Academy given that Sampson High School had similar student demographics and was located in the same school district. Further, the majority of students across the nation attend comprehensive high schools and we wanted to compare a traditional student experience to students’ experiences in a career academy (a comprehensive high school reform initiative) that focuses on a career-theme with a smaller learning community model. We hypothesized that students in a career academy would benefit from lower occurrences of bullying in the school.

Instrumentation of the survey of high school student engagement

Instrument validation

We established validity of the high school student engagement instrument. First, to establish content validity, we modified the original high school survey of student engagement instrument to fit key elements of the academy model and comprehensive school. Second, we sent the modified instrument to a panel of experts. The six-person expert panel represented researchers with expertise in the evaluation of career academies as well as high school administrators. The panel of experts evaluated the latent variables of the instrument for their level of appropriateness. Third, based on the recommendations from the expert panel, we clarified, modified, and added additional items as necessary. Fourth, we conducted cognitive interviews with six high school students to ensure the questionnaire items were readable. Fifth, we conducted a pilot study with 30 students.

Survey dimensions

For the purposes of this study we focused on bullying behavior—both experienced and observed—items from the questionnaire as recommended in the bullying literature (Mehta et al., 2013). The item that measured students’ experience being bullied asked: “during this school year, how often have you been picked on or bullied by another student.” The item that measured students’ observations of bullying asked: “during this school year, how often have you witnessed an act of bullying.” Both items used a 4-point Likert-type scale: 1 = “Never,” 2 = “Rarely,” 3 = “Sometimes,” and 4 = “Often.” Both of these items were then recoded into dichotomous variables combining “Rarely,” “Sometimes,” and “Often” into “yes” for having experienced or observed bullying.

Data analysis

To reduce the effect of the initial differences between samples, we used propensity score matching to match students from NAF academy schools with similar students from the comprehensive high school (Luellen, Shadish, & Clark, 2005; Rubin, 1997; Rubin & Thomas, 2000). Propensity score matching helps in reducing bias associated with confounding variables that could explain the outcome instead of or in addition to the treatment. This also assists in reducing bias stemming from our non-ability to conduct a randomized experiment. We attempted to control for the differences in the NAF students and comprehensive students by controlling for differences in gender (male/female/nonconforming), ethnic and racial background (American Indian/Alaska Native; Asian or Asian American; Native Hawaiian or Other Pacific Islander; African American or Black; Hispanic, Latinx, or Spanish origin; White; Other; Prefer not to respond; and Multiracial), socio-economic status (measured by free/reduced lunch status), grade in school (9^th through 12^th), and language primarily spoken at home (English, Spanish, or other). We created propensity scores for each student based on the aforementioned characteristics. We then used the nearest neighbor matching function—without replacement—to pair each participant or non-participant on each individual’s propensity scores.

Using the pairs found in the propensity score matching, we compared group differences using t-tests for the two items on bullying. Because statistical significance is sensitive to sample size, Cohen’s d effect sizes (mean difference/pooled standard deviation) were also calculated and considered in this study (Cohen, 1988, 1992). Cohen (1988) reluctantly defined effect sizes as "small, d= .2," "medium, d = .5," and "large, d = .8," but also cautioned that "there is a certain risk inherent in offering conventional operational definitions for those terms used in power analysis in as diverse a field of inquiry as behavioral science" (p. 25) and urged researchers to use their discipline’s context when interpreting effect sizes. Within the educational context, some researchers have concluded that these cut offs may be rather ambitious (Gonyea & Sarraf, 2009; Mayhew et. al, 2016). For example, Mayhew et al. (2016) recommended lower cut offs in higher education research, where Cohen’s d effect size of less than .15 represents a trivial difference, .3 represents a medium difference, and.5 to represents a large difference. Still, to be cautious and to not overemphasize small differences, we used the stricter guidelines set by Cohen (1988).

Finally, we produced two logistic regression models to determine if differences exist between academy students and their peers at the comprehensive high school within the area of bullying—even when controlling for other factors. The two dependent variables were the dichotomized variables for students’ experiences with and observations of bullying. The independent variables included five student demographic characteristics: gender, ethnic and racial backgrounds, free/reduced lunch, grade in school, and primary language spoken at home. All categorical independent variables were dummy coded prior to entry into the model.

Results

Prior to estimating the relationships between academy participation and the two bullying items, we created the probit regression model predicting participation in an academy. We then used those results to create both inverse probability weights and assess the sample’s balance with and without those weights. We determined that applying the weights resulted in minimal differences in the observable characteristics between students attending academies compared with their comprehensive high school peers. Prior to weighting, the mean absolute standardized difference was .07, and when weighted it was .03. After concluding that our sample met the balancing assumptions of the propensity score model, we ran the two independent t-tests. Additionally, for the purposes of simplicity and the ability to explain in more traditional ways, we used the nearest neighbor function without replacement to match each of the students in the academies (treated) with someone at the comprehensive highs schools (untreated) based on their propensity scores. We used nearest neighbor matching because it is the most straightforward matching estimator (Caliendo & Kopeinig, 2008). The propensity score matching resulted in 299 matched pairs with no missing data for the variables included in any of the analyses for this study. We used the pairs to compare the demographic characteristics of students in the academy and the comprehensive school (See Tables 1 and 2). As would be expected, no statistically significant differences existed between the two groups of matched pairs.

Table 1. Summary of participants’ demographic background by school type for the initial sample.

	Career academy (N = 300)	Comprehensive school (N = 730)	χ^2	P
Gender			93.26	***
Female	25.4% (76)	58.3% (285)
Male	71.9% (215)	39.3% (423)
Nonconforming	2.7% (8)	2.3% (17)
Ethnic and racial background
AmerInd/Alaskan	0.3% (1)	0.8% (6)	48.69	***
Asian or Asian American	7.0% (21)	15.4% (112)
African American/Black	8.7% (26)	19.3% (140)
Hawaiian/Pacific Islander	0.3% (1)	0.7% (5)
Latinx	12.0% (36)	9.0% (65)
White	44.1% (132)	27.6% (200)
Multi-racial	23.7% (71)	22.9% (166)
Other	1.0% (3)	1.9% (14)
Prefer not to respond	2.7% (8)	2.3% (17)
Grade level			10.64	**
9th	38.1% (114)	36.7% (268)
10th	20.7% (62)	29.2% (213)
11th	24.1% (72)	18.1% (132)
12th	17.1% (51)	16.0% (117)
Free/reduced price lunch			8.42	*
No	36.1% (108)	36.0% (261)
Yes	32.1% (96)	38.6% (280)
Don’t know	29.1% (87)	21.5% (156)
Prefer not to respond	2.7% (8)	3.9% (28)

*p <.05; **p <.01; *** p <.001

Table 2. Summary of participants’ demographic background by school type for the matched pair sample.

	Career academies (N = 299)	Comprehensive school (N = 299)	χ^2	P
Gender			1.22	NS
Female	25.4% (76)	29.4% (88)
Male	71.9% (215)	67.9% (203)
Nonconforming	2.7% (8)	2.7% (8)
Race/ethnicity			3.48	NS
Amerind/Alaskan	.3% (1)	.7% (2)
Asian or Asian American	7.0% (21)	6.0% (18)
Black or African American	8.7% (26)	11.0% (33)
Hawaiian/Pacific Islander	.3% (1)	.0% (0)
Latinx	12.0% (36)	13.7% (41)
White	44.1% (132)	40.5% (121)
Multi-racial	23.7% (71)	24.4% (73)
Other	1.0% (3)	.7% (2)
PNR	2.7% (8)	3.0% (9)
Grade			2.59	NS
9th	38.1% (114)	40.1% (120)
10th	20.7% (62)	22.4% (67)
11th	24.1% (72)	18.7% (56)
12th	17.1% (51)	18.7% (56)
Free/reduced price lunch			2.32	NS
No	36.1% (108)	32.8% (98)
Yes	32.1% (96)	37.5% (112)
Don’t know	29.1% (87)	26.4% (79)
Pnr	2.7% (8)	3.3% (10)

In order to determine the probability that our research design will detect the effects of our treatment (participation in NAF), we conducted post-hoc power analyses. With the 299 matched pairs, the study has 80% power to detect an effect size of .23 or greater when comparing means. Since the minimum effect size that we would consider meaningful and not trivial would be 0.30 (Albers & Lakens, 2018; Cohen, 1988; Mayhew et. al, 2016), the power analyses indicated that our sample size would be sufficient to detect significant effects for mean comparisons. For the regression models, using the number of predictors and R² (discussed later) for the two models, the observed statistical power were 0.99990148 for Experienced Bullying and 0.99999997 for Observed Bullying. These power levels are well above the .80 standard (Cohen, 1992) and suggest minimal concerns for a Type I error occurring.

Response to research questions

Results indicated that the academy students, when compared to comprehensive school students, both experienced lower levels of bullying (M = .27 and M = .46, respectively) and observed lower levels of bullying (M = .40 and M = .69, respectively). Additionally, these differences were statistically significant at the p < .001 level. Finally, the effect sizes suggest that these are nontrivial differences and would be considered near medium for the experienced bullying (d = −.41) and medium for the observed bullying (d = −.59) —even by Cohen’s conservative standards (Cohen, 1998). For more specifics, see Table 3 below.

Table 3. Summary of T-tests and effect sizes comparing academy students with comprehensive students on bullying items for both samples.

	Academy (n = 299)	Comp. (n = 299)	Mean Difference	t	Effect Size d
Experienced Bullying	.27	.46	−.19	5.01***	−.41
Observed Bullying	.40	.69	−.29	7.18***	−.59

*p <.05; ** p <.01; *** p <.001

As presented in Table 4, our logistic regression results indicated that participation in an academy was significantly related to the amount of bullying—both experienced and observed—that high school students encounter, even after controlling for students’ gender, ethnic and racial backgrounds, free/reduced lunch, grade in school, and primary language spoken at home. We also found statistically significant differences in both the students’ perceived experiences around bullying and their observations of others being bullied. NAF academy students were more than half as likely to report experiences with bullying than their counterparts attending a comprehensive high school (Odds ratio = 0.43; p<.001), and less than one-third as likely to report observing others being bullied (Odds ratio = 0.29; p < .001). Since the school type (Academies in Table 4) coefficients in both of these models were negative, calculating the inverse odds ratios maybe easier to interpret. In other words, students at comprehensive high schools were almost two and half times as likely to report experiencing bullying (inverse Odds ratio = 2.33) and three and half times as likely to report observing bullying (inverse Odds ratio = 3.45). For the experienced bullying model, the statistically significant demographic characteristics related to bullying were gender, ethnicity/race, and language spoken at home. Gender nonconforming students reported higher levels of experiences with being bullied (Odds ratio = 3.00; p < .05) than their male counterparts and Asian students experienced nearly one-fourth of the bullying (Odds ratio = 0.26; p < .01) when compared to their White peers. Additionally, students who spoke primarily Spanish at home were half as likely to report being bullied than those whose primary language spoken at home was English (Odds ratio 0.54; p < .05). For the observing others being bullied model, only grade level was statistically significant demographic. Those in either 11^th (Odds ratio = 2.38; p < .001) or 12^th (Odds ratio = 1.67; p < .05) grade had a higher likelihood of reporting that they observed bullying than their 9^th grade counterparts.

Table 4. Logistic regression models for two bullying items (n = 299).

	Experienced bullying				Observed bullying
	B	SE	Odds ratio	p	B	SE	Odds ratio	p
School^a
Academies	−.86	.180	0.43	***	−1.24	.179	0.29	***
Gender^b
Female	.21	.204	1.24		.23	.205	1.26
Nonconforming	1.10	.563	3.00	*	1.03	.638	2.80
Grade^c
10th	.29	.242	1.34		.27	.240	1.31
11th	.40	.242	1.48		.87	.247	2.38	***
12th	−.03	.264	0.97		.51	.258	1.67	*
Ethnicity/Race^d
AmerInd/Alaskan	−.28	1.257	0.75		−1.13	1.276	0.32
Asian	−1.34	.504	0.26	**	−.46	.432	0.63
Black	−.15	.322	0.86		−.06	.326	0.94
Latinx	.08	.393	1.08		.06	.385	1.06
Multi-Racial	.20	.251	1.22		−.03	.255	0.97
Other	−.18	1.061	0.84		1.51	1.204	4.54
Prefer not resp.	.11	.555	1.11		.22	.584	1.24
Language spoken at home^e
Spanish	−.61	.312	0.54	*	−.06	.303	0.95
Other	.29	.307	1.33		.53	.315	1.69
Free/reduced price lunch^f
Yes	.27	.222	1.31		.41	.224	1.51
Don’t know	−.20	.237	0.82		−.19	.229	0.83
Prefer not resp.	−.19	.561	0.83		−.15	.534	0.86

For experienced bullying: Cox and Snell pseudoR-squared 0.09; Nagelkerke pseudoR-squared 0.12; 64.2% correctly classified

For observed bullying: Cox and Snell pseudoR-squared 0.13; Nagelkerke pseudoR-squared 0.17; 65.4% correctly classified

Reference group:

^aComprehensive

^bReference group: Male

^cReference group: 9th

^dReference group: White

^eReference group: English

^fReference group: No

*p <.05; ** p <.01; *** p <.001

For a robustness check, we also ran logistic regression models with the full sample, which contained 1,162 respondents representing 317 career academy and 845 comprehensive students. Of the 317 career academy students, 18 were missing on one or more of the demographic variables included in the propensity score matching process. Of the 845 comprehensive students, 120 were missing on one or more of the demographic variables. Typically these cases were missing on all the demographics because they did not complete the entire survey and demographics were found at the end of the survey. None of the 299 matched pairs were missing data on items in analyses. We wanted to ensure that the small counts found in some of the subgroups (such as in the race/ethnicity categories) were not causing any nonsignificant findings or Type I errors. Results for these models mirrored the results found with the 299 match pairs. The only significant finding that was added for these full sample models was for the group of “Gender: Nonconforming” in the observed bullying model (Odds ratio = 2.38; p < .05), already found to be statistically significant in the experienced bullying model. However, we acknowledge the small sample size of nonconforming respondents in this study and thus readers should use caution in making interpretations. For the regression model details see the Appendix.

Discussion

In this study, we examined the role that school type—a magnet career academy and a large comprehensive school—plays in the likelihood of students experiencing and observing bullying. Researchers have pointed out that students in urban schools—with high percentages of low-income students—are more likely to experience school violence (Lleras, 2008). However, studies have not investigated how school violence—particularly bullying—differs across urban school contexts as it relates to small magnet career academies compared to large, traditional comprehensive high schools. Given the growing popularity of career academies across the nation (MDRC., 2020), we believe it is critical to examine how career academies fare in regard to bullying compared to students at a large, traditional comprehensive school. To that end, we found that academy students, when compared to comprehensive school students, both experienced and observed significantly lower levels of bullying.

It was not surprising that academy students reported lower levels of experiencing and observing bullying given related literature focusing on the student experience in career academies. For example, Fletcher and Cox (2012) found that African American academy students believed their participation in a career academy was beneficial as it created a sense of community within their school. These academy students perceived their peers and teachers as an extension of their families. Similarly, based on findings from stakeholders of an academy, Fletcher et al. (2019) described career academies as laboratories of equity, safety, and inclusion for students. They attributed these characteristics of the academy to key elements of the academy model, namely: the open enrollment admissions policy, the nature of a small school, and the shared interest in information technology with peers. Even further, Fletcher et al. (2020) found that academy students were significantly more likely to be emotionally engaged—sense of safety and belonging to their school—compared to those in a large, traditional comprehensive high school. These are all key elements that are likely to promote a positive school culture and climate serving as a buffer, moderator, and mediator to school violence. Similarly, in this study, we believe that the lower rates of experiencing and observing bullying are likely related to the small size of Cascade Academy enabling students to have a more personalized schooling experience with school personnel. This notion supports prior researchers that found lower rates of bullying in smaller schools (Lleras, 2008; Cornell et al., 2013), while also adds disconfirming evidence to researchers that found actual lower rates of bullying in larger schools (Klein & Cornell, 2010) as well as those who found no relationship between bullying and school size (Khoury-Kassabri et al., 2004; Olweus, 1995). Further, the perceptions of the students at Cascade Academy and Sampson High School seem to align with the rates of disciplinary actions within their respective schools − 20% compared to 30%, respectively.

In addition, we believe that students’ shared interest in the IT career theme with peers adds to the camaraderie and sense of belonging and community. In this regard, it is important to note that Cascade Academy is a magnet school where students make a choice to apply for admission. Cascade Academy administrators rely on open enrollment –meaning that they do not select students based on prior achievement or other academic measures. Instead, students are selected based on a lottery system. This finding adds to the research that has demonstrated positive student outcomes for magnet schools (Ballou et al., 2006; Bifulco et al., 2008; Gamoran, 1996; Poppell & Hague, 2001). Our research findings also support the idea of establishing whole school magnets as opposed to school-within-a-school programs that have been known to racially separate students within the same school (Frankenberg & Siegel-Hawley, 2008). In total, it is likely that students at Cascade Academy benefit from a combination of factors, including their abilities to choose to participate in the academy, having peers with similar career interests, and the small size of the academy. These factors are strong predictors of a positive school culture and climate that lends itself to lower incidences of school violence, particularly bullying (Fletcher et al., 2019).

While our findings show no significant differences between males and females as it relates to bullying, other studies have consistently reported that boys report more bullying compared to girls (Espelage & Holt, 2001; Nansel et al., 2001; Scheithauer, Hayer, Petermann, & Jugert, 2006; Wei et al., 2010). Studies have also demonstrated that ethnically and racially diverse students often experience bullying (Sawyer, Bradshaw, & O’Brennan, 2008). In addition, we recommend that future studies identify non-dominant students in their studies, such as non-gender conforming, LGBT, and students with disabilities. We recommend the identification of these individuals to provide a glimpse into their school experiences related to school violence, particularly bullying. In terms of practical implications for schools, we recommend that schools emphasize the recruitment of positive role models and leaders (e.g., teachers, administrators, mentors, guest speakers) from non-dominant groups to expose their students to individuals from an array of backgrounds and cultures. Such strategies are often complemented with the promotion of discussions regarding social justice issues—including issues such as racism and homophobia (Buhs et al., 2006).

This study adds to the growing body of literature examining what contributes to school bullying—given the paucity of research on school characteristics factors that play a role in its pervasiveness. More specifically, we believe findings from this study adds insight into the question of why students in some studies report higher rates of observed and experienced bullying in certain types of schools. Interestingly, apart from non-gender conforming students, we found that bullying is not significantly different based on student demographics and grade level. Instead, our study underscores the notion that school type—magnet career academy compared to large, traditional comprehensive school—is a significant factor in predicting bullying. With this frame of reference, we recommend that school administrators ideally attempt to provide students with the ability to choose to participate in a school community (possibly a small learning community for non-magnet schools), thereby, having students with peers that share similar career or affinity interests and reducing the large feeling of a comprehensive high school. We believe that these changes can help schools create a more positive school culture and climate that lends itself to lower incidences of school violence, particularly bullying. Hence, we hope that by accumulating knowledge on school bullying based on school type we can reduce violence and create a healthy and caring school environment for all students.

Limitations

However, it is also important for readers to use caution in generalizing our findings to other schools and settings given the focus on only two schools and specific types of schools—small magnet academy and large comprehensive. We also are unable to decipher the influence of each school characteristic (e.g., magnet, career academy, small size) on bullying. It is also unknown whether the decrease in bullying could be partially attributed to indirect effects (e.g., achievement, student engagement). In addition, we only relied on two measures (e.g., observations of bullying and victimization of bullying) instead of a more robust number of items that could examine bullying as a multidimensional construct that captures the three components of Olweus definition. Further, the measures in our analyses relied solely on self-reported data from students in the two schools. Therefore, the extent that our findings are generalizable depends on the degree that schools share similar characteristics (e.g., magnet schools that use a lottery system for student admission and comprehensive schools with large student populations).

Additional information

Funding

This research was supported by a grant from the National Science Foundation (Grant # 1614707 & 2016580).

Notes on contributors

Edward C. Fletcher

Edward C. Fletcher Jr., Workforce Development and Education, The Ohio State University.

Amber D. Dumford

Amber D. Dumford, Department of Leadership, Policy and Lifelong Learning, University of South Florida.

Appendix.

Logistic regression models for two bullying items with full sample (n = 317 for career academy & n = 845 for comprehensive students)

Table

	Experienced Bullying				Observed Bullying
	B	SE	Odds ratio	p	B	SE	Odds ratio	p
School^a
Academies	−.95	.163	0.39	***	−1.20	.158	0.30	***
Gender^b
Female	.23	.141	1.26		.21	.144	1.24
Nonconforming	.93	.449	2.54	*	1.05	.535	2.85	*
Grade^c
10th	.08	.168	1.08		.01	.171	1.01
11th	.23	.186	1.26		.54	.196	1.72	**
12th	.07	.200	1.07		.31	.204	1.36
Ethnicity/Race^d
AmerInd/Alaskan	−.76	.855	0.47		−.80	.796	0.45
Asian	−.67	.257	0.51	**	−.46	.258	0.63
Black	−.25	.210	0.78		−.22	.217	0.81
Latinx	−.25	.311	0.78		.08	.318	1.08
Multi-Racial	.20	.194	1.22		.10	.202	1.11
Other	.41	.532	1.50		1.00	.681	2.72
Prefer Not Resp.	.03	.455	1.03		.24	.494	1.27
Language Spoken At Home^e
Spanish	−.27	.232	0.77		−.03	.241	0.97
Other	.15	.220	1.16		.32	.230	1.38
Free/Reduced Price Lunch^f
Yes	.18	.165	1.19		.26	.170	1.30
Don’t know	−.11	.180	0.90		.02	.181	1.02
Prefer Not Resp.	.09	.387	1.09		.40	.415	1.49

For experienced bullying: Cox and Snell pseudoR-squared 0.07; Nagelkerke pseudoR-squared 0.10; 60.4% correctly classified For observed bullying: Cox and Snell pseudoR-squared 0.10; Nagelkerke pseudoR-squared 0.13; 66.5% correctly classified

Reference group:

^aComprehensive

^bReference group: Male

^cReference group: 9th

^dReference group: White

^eReference group: English

^fReference group: No

*p <.05; ** p <.01; *** p <.001