© 2014 American Psychological Association 1045-3830/14/$ 12.00 http://dx.doi.org/10.1037/spq0000072
School Psychology Quarterly 2014, Vol. 29, No. 3, 287-305
Teacher and Staff Perceptions of School Environment as Predictors of Student Aggression, Victimization, and Willingness to Intervene in Bullying Situations Dorothy L. Espelage
Joshua R. Polanin
University of Illinois at Champaign-Urbana
Vanderbilt University
Sabina K. Low Arizona State University This study examines how teacher and staff perceptions of the school environment correlate with student self-reports of bullying, aggression, victimization, and willing ness to intervene in bullying incidents using multi-informant, multilevel modeling. Data were derived from 3,616 6th grade students across 36 middle schools in the Midwest, who completed survey measures of bullying, aggression, victimization, and willingness to intervene in bullying situations. Teachers and staff (n = 1,447) com pleted a school environment survey. Bivariate associations between school-level and student self-reports indicated that as teacher and staff perceive aggression as a problem in their school, students reported greater bully perpetration, fighting, peer victimization, and less willingness to intervene. Further, as staff and teacher report greater commit ment to prevent bullying and viewed positive teacher and student relationships, there was less bullying, fighting, and peer victimization, and greater willingness to intervene. In a model where all school environment scales were entered together, a school commitment to prevent bullying was associated with less bullying, fighting, and peer victimization. Student-reports of bully perpetration and peer victimization were largely explained by staff and teacher commitment to bully prevention, whereas fighting and willingness to intervene were largely explained by student characteristics (e.g., gender). We conclude that efforts to address bullying and victimization should involve support from the school administration. School psychologists should play an active role in the school climate improvement process, by creating a school climate council consisting of students, parents, and teachers; administering school climate measures; identifying specific school improvement targets from these data, and engaging all stakeholders in the ongoing school improvement plan. Keywords: bullying, school climate, multilevel modeling, middle school, teacher
Com prehensive school-w ide efforts to pre vent aggression, bullying, and victim ization are predicated on the assum ption that reducing these behaviors m ust include w orking w ith all stakeholders in the school com m unity, includ
ing teachers, students, and parents (Espelage, in press; O lw eus & L im ber, 2010). D raw ing from the ecological m odel o f child developm ent, the school environm ent is an im portant m icrosys tem that influences how students’ view bullying
This article was published Online First August 4, 2014. Dorothy L. Espelage, Department of Educational Psy chology, University of Illinois at Urbana-Champaign; Joshua R. Polanin, Peabody Research Institute, Vanderbilt University; Sabina K. Low, School of Social and Family Dynamics, Arizona State University. Research for the current study was supported by the Centers for Disease Control & Prevention (1U01/ CE001677) to Dorothy L. Espelage (PI) at the Univer
sity of Illinois at Urbana-Champaign. Opinions ex pressed herein do not necessarily reflect those of the Centers for Disease Control & Prevention, or related offices within. Correspondence concerning this article should be ad dressed to Dorothy L. Espelage, Department of Educa tional Psychology, University of Illinois at UrbanaChampaign, Champaign, IL 61820-6925. E-mail: espelage @illinois.edu 287
288
ESPELAGE, POLANIN, AND LOW
and aggression, how adult role models correlate with student behavior, and how school-level norms and policies correlate with these student behaviors (Bronfenbrenner, 1977; Espelage, in press). Studies consistently report that negative school environmental factors (e.g., policies, staff reaction to bullying) can lead to an in crease in the frequency of bullying, aggression, and victimization and reduce the likelihood of students feeling safe in their school (Espelage, Bosworth, & Simon, 2000; Goldweber, Waasdorp, & Bradshaw, 2013). In contrast, youth with positive perceptions of their school envi ronment are less likely to exhibit externalizing behaviors (e.g., aggression; Espelage et al., 2000; Goldweber et al., 2013; Totura et al., 2009). The majority of these studies, however, are limited to student self-report of school cli mate or a narrow assessment of school environ ment (e.g., safety, belonging). The current study addresses this gap in the literature by examining teacher and staff perceptions of the school en vironment and how these perceptions correlate with student self-reports of bullying, aggres sion, victimization, and willingness to intervene in bullying incidents. To examine the mac rolevel correlations with student behavior that move beyond the classroom level, we utilize a broader, multidimensional assessment of school environment to include a comprehensive focus and commitment to bully prevention and sup ports to teacher and staff to learn ways to stop bullying. In addition, we use multilevel model ing to capture precisely student behaviors as a function of the larger social context at the school-level. School Environment and Climate Predictors School environment is a broad term that en compasses multiple features of school climate or “culture,” and in this article, refers to the psychosocial quality and character of school life (i.e., we did not assess physical qualities of the school; G ottfredson, Gottfredson, Payne, & Gottfredson, 2005). School climate is based on patterns of people’s experience of school life and reflects norms, goals, values, interpersonal relationships, teaching, learn ing, leadership practices, and organizational structures (National School Climate Council, 2007). Because of frequent interactions be
tween teachers and students in school, it is not surprising that teacher’s and other staff’s per ceptions of the environment are associated with student attitudes and behaviors. In a study of 40 countries, Harel-Fisch and col leagues (2011) analyzed the World Health Organization health behavior in school-age children (WHO-HBSC) surveys and found that as negative school perceptions reported by students increased, so did their involve ment in bullying as a perpetrator or victim. Building on these findings, in the current study we examine how six indices of school climate and environment from the teacherstaff perspective correlate with bullying, vic timization, aggression, and willingness to in tervene in bullying situations from student report. The school climate and environment scales includes perceptions of (a) staff and student intervention in bullying, (b) aggres sion as a problem, (c) administrative support and opportunities to prevent bullying and vi olence, (d) positive teacher-staff-student rela tions, and (e) gender equity and intolerance of sexual harassment. Staff and Student Intervention Research indicates that bully perpetration and victimization rates are higher and willingness to intervene is lower when students perceive adults’ prevention and intervention efforts as ineffective (Goldweber et al., 2013; Waasdorp, Pas, O’Breenan, & Bradshaw, 2011). Therefore, we hypothesized that greater levels of teacher and staff intervention would be associated with less student reports of bullying, victimization, aggression, and greater willingness to intervene. Aggression as a Problem It has been noted that there are discrepancies between how teachers and staff perceive bully ing in comparison to their students. Many teach ers are unaware of how serious and extensive the bullying is within their schools, and are often ineffective in being able to identify bul lying incidents (Bradshaw, Sawyer, & O’Brennan, 2007; Waasdorp et al., 2011). Di vergence between staff and student estimates of the rates of bullying are seen in elementary, middle, and high school, with staff consistently underestimating the frequency of these events (Bradshaw et al., 2007). In a large-scale valida-
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
tion study of three climate measures, Bandy opadhyay and colleagues (2009) found signifi cant, albeit modest, correlations between teacher and student reports of bullying and teas ing across 291 Virginia high schools. Thus, we hypothesize that teacher and staff perceptions of aggression at the school-level will be signifi cantly associated with student reports of bully ing, fighting, and victimization.
sexual harassment or treat boys and girls differ ently, students report greater peer victimization and less willingness to seek help for themselves or other peers (Charmaraman, Jones, Stein, & Espelage, 2013). Therefore, it was hypothesized that lower bullying and victimization rates and greater willingness to intervene would be asso ciated with greater gender equity or intolerance of sexual harassment at the teacher-staff level.
School Commitment to Bully Prevention Orpinas and Home (2006) argue that bully occurs less often in schools where the leader ship provided by the school administration works closely with staff to understand policies and seek input from all adults. Twemlow and Sacco (2013) argue that bullying is a universal dysfunctional social process, where bullying and victimization at the student level are symp toms of an adult culture that are not attending to the issue at hand. When adults deny the prob lem, youth report greater aggression and victim ization. Thus, we hypothesized that selfreported student levels of bullying, aggression, and victimization would be lower in schools where teachers and staff perceive that prevent ing bullying and violence is a priority and they are provided with professional development op portunities. Positive Teacher-Staff-Student Interactions Youth with lower levels of school connect edness were significantly more likely to be in volved in bullying and peer victimization (Espelage et al., 2000; Glew et al., 2005; Goldweber et al., 2013). Thus, we hypothesized that closer teacher-student relations reported by the teachers and staff would be associated with less bullying, aggression, victimization, and greater willingness to intervene by students.
289
Current Study In summary, the extant research literature suggests that the school environment is associ ated with student’s self-reported bullying, ag gression, victimization, and student’s willing ness to intervene to help other students. However, only a few studies have used multi level analyses to fully capture the associations of school climate on student attitudes and be haviors, but these studies are limited. Waasdorp and colleagues (2011), for example, used mul tilevel modeling with student, staff, and parent data, but used single items to assess school safety, belonging, witnessing bullying, and bully perpetration and victimization. Richard and colleagues (2012) surveyed over 18,000 French youth and examined school-level char acteristics using multilevel modeling largely from aggregated student reports because only 701 teachers completed measures. This current study adds to the growing body of literature on bullying and school climate by assessing per ceptions of school environment from teachers and staff across multiple scales and selfreported behaviors from students. Furthermore, the teacher and staff environment surveys were completed by teachers, paraprofessionals, ad ministrators, and other adults in each school, offering a broader assessment of the school environment.
Gender Equity and Intolerance of Sexual Harassment
Method Participants
Verbal aggression, peer victimization, or bul lying during early adolescence can involve ho mophobic name-calling and sexual commentary or sexual touching, and when these are un wanted they are referred to as sexual harassment or sexual violence (Espelage, Basile, & Ham burger, 2012; Poteat & Espelage, 2007). When the adults in middle school are dismissive of
In total, 3,616 6th grade students across Illi nois and Kansas middle schools participated in the surveys (see Table 1). The students ranged in age from 11 to 13 years old, but most stu dents were 11 (Illinois = 75.8%, Kansas = 78.4%). A slight majority of students were male in both Illinois and Kansas (Illinois = 51.2%,
290
ESPELAGE, POLANIN, AND LOW
Table 1 Participant and School Characteristics Illinois Number of students Number of schools Age 11 12 13 Gender Male Female Race Black Asian Latina/o White Biracial Mother’s education Less than high school High school graduate Some college Graduate from college Some or all postgraduate Unknown/no response Father’s education Less than high school High school graduate Some college Graduate from college Some or all postgraduate Unknown/no response Teacher and staff N Average N per school Median N per school Percent free or reduce lunch Percent females Percent White Percent Black
2,012 24
Kansas 1,604 12
75.8 22.5 1.7
78.4 20.7 0.9
51.2 48.8
52.9 47.1
33.5 1.0 33.1 20.2 11.7
17.5 4.2 35.2 30.4 12.8
17.9 25.8 19.0 20.2 9.9 7.2
13.3 19.0 15.4 15.7 7.3 29.2
19.2 25.9 15.4 17.1 9.8 13.6 732 30.08 27.50 71.95 49.4 19.4 43.5
13.8 19.3 10.6 13.0 6.0 37.3 715 59.08 58.50 78.52 42.8 33.0 19.1
Kansas = 52.9%). In Illinois, Blacks (33.5%) and Latino/as (33.1%) constituted the majority of students surveyed. Most students in Kansas schools were Latino/a (35.2%) or White (30. 4%). Overall, 73% of students were eligible for free or reduced lunch (FRL) across both states. States did not differ significantly with regard to the percentage of FRL children or percentage of females (Illinois = 71.9, Kansas = 78.9). Illi nois students indicated that the majority of their mothers graduated from college (20.2%) com pared with mothers of Kansas students (15.7%). A large portion of students from Kansas, how ever, either did not know their mother’s educa tion or left this item blank.
The analyses presented here include baseline data from a large-scale randomized clinical trial of a social-emotional learning program that is described in more detail elsewhere (Espelage, Low, Polanin, & Brown, 2013). The sample, therefore, was drawn initially to test the efficacy of this program in a nested cohort (6th graders) longitudinal study. Schools were recruited through the school district offices and had to be willing to be assigned to either an intervention or control condition. To be eligible, the schools could not be implementing any large-scale bully prevention curriculum or initiative, and control schools agreed to not implement a bully preven tion program until after the 3 year trial. Twentyfour schools from Illinois and 12 schools from Kansas participated in the project. All staff were invited to complete the survey, not just 6th grade teachers. In Illinois schools, 732 teachers and other staff completed the survey and 715 teachers and other staff from Kansas schools participated. One school in Illinois provided only two teacher and staff surveys; all other schools, across both states, had at least 10 teach ers participate in the survey (range = 2-101; M = 39.69; SD = 24.24). Schools were eligible for the school-level stipend if 80% of their teachers and staff completed the survey; all but three schools reached this goal. The sample consisted of 66% teachers, 10% support staff, 9% paraprofessionals, 4% administrators, 3% counselors or psychologists, 2% custodian staff, and 1% cafeteria staff. The staff identified mainly as White (75%), 10% as Black, and 8% as Hispanic, and 78% of the sample identified as female. The average age of the sample was 42.5 years old. Data Collection A waiver of active parental consent and an active consent protocol were both approved by the institutional review board at the University of Illinois, and districts could use either method. Parents of all 6th grade students enrolled in all participating schools were sent letters or con sent forms. An 86% participation rate was achieved in schools using a waiver of active consent and 63% participation rate was achieved for schools using an active consent procedure. Students were asked to consent to participate in the study through an assent pro cedure included on the coversheet of the survey.
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
At each wave of data collection, six trained research assistants, the primary researcher, and a faculty member collected the data. At least two of these individuals administered surveys to classes ranging in size from 10 to 25 students. The research assistants first informed students about the general nature of the investigation. Students were then given survey packets and the survey was read aloud to them. It took students ~ 40 min to complete the survey. Teachers and staff at each school were sent an email with a link to the online School Environment Survey. Teacher and Staff School Environment Survey Data were collected from school staff mem bers using the School Environment Survey. This instrument was adapted from the Colorado Trust’s Bullying Prevention Initiative (Csuti, 2008) for an elementary school study. Demo graphic information related to staff members’ gender, race or ethnicity, age, position in the school, and length of employment at the school is included in the survey. Because the current study was conducted in middle schools and this measure was first validated with elementary schools, these items were subjected to factor analysis to determine the scale score. To deter mine the latent factor structure of the school environment survey variables, an exploratory factor analysis (EFA) was utilized (Kahn, 2006). The EFA followed the procedures out lined in Kline (2011). Concern was given to the number of response categories utilized, but Kline (2011) indicated that four responses were reasonable for maximum-likelihood estimation procedures. A principal axis factoring was uti lized to extract the factors and a varimax rota tion was implemented to rotate the structure. A scree plot and subsequent parallel analysis (Hayton, Allen, & Scarpello, 2004) confirmed the latent factor structure and the final model was fixed to extract the exact number of factors required (8). Items that loaded above .3 on the hypothesized factor were retained (Kline, 2011); items that failed to meet these criteria were subsequently dropped from the variable construction. We did not consider the clustered nature of the data for the factor analysis because the statistical significance of the factor loadings was not considered as important as the factor loadings and because the latent factors were
291
aggregated to the school-level (Stapleton, 2006). We conducted the factor analysis in SPSS version 21 (IBM Corp., 2013); we uti lized O’connor’s (2000) SPSS macro to conduct the parallel analysis. Results of the principal axis factor explor atory factor analysis revealed nine potential fac tors. Upon further review of the scree plot and parallel analysis, however, only eight factors were perceived as viable. Six of the eight fac tors identified were used in this manuscript. The two other factors were considered not theoreti cally important and, therefore, dropped from the analyses. The six factors’ item loadings, eigen values, percentage of variance, and Cronbach’s a coefficients are detailed in Table 2. The Cronbach's a coefficients ranged from .79 to .94. Overall, the six factors accounted for 56.8% of the total variance. Student intervention. Five items emerged in this first factor. Teachers and staff are asked “How likely is it that STUDENTS at your school could be counted on to help out in the following situations?” Examples include (1) A student is making fun o f and teasing another student who is obviously weaker, (2) A student is spreading rumors or lies about another stu dent behind their back. Response options in clude “Very unlikely,” “Unlikely,” “Likely,” and “Very likely.” A Cronbach’s a coefficient of .84 was calculated for this sample of teachers and staff. Staff intervention. Five items emerged in this second factor. Teachers and staff are asked “How likely is it that STAFF at your school could be counted on to help out in the following situations?” And the same five items from the student intervention (above) are presented. Re sponse options include “Very u n l i k e l y “Un likely,” “Likely,” and “Very likely.” A Cron bach’s a coefficient of .91 was calculated for this sample of teachers and staff. Perceptions of aggression as a problem. Five items emerged in this third factor. Teach ers and staff are asked “How much of a problem are the following issues at your school?” Exam ples include (1) Students picking fights with other students', (2) Students who push, shove, or trip weaker students. Response options include “Not a problem,” “Sort o f a problem,” “Pretty big problem,” and “Huge problem.” A Cron bach’s a coefficient of .87 was calculated for this sample of teachers and staff.
ESPELAGE, POLANIN, AND LOW
292
CT O d c 2 cn R
Standardized Factor Loadings, Eigen Values, % Variance, and Cronbach’s a Coefficients for Teacher and Staff Variables
Table 2
o •
c
c
3 o .5 CQ " s I o
3 2 X 0 00 0 X £ 33 0c C/5 rQ ^ 03 c = x: y .3 -e o .s § M 03 :■ s-o ,c :5 *3 • — 03 o 3 . -P M -a - ^ o> a2 -c : 3 -o 00 "2 3 < £ -c 3 o c/3 3 3 3 o c 35 ^ . C a/5- SO £ o c/5 T3 . C C/5 i & cd . C/5 CO >o o
•a
17. D e m o n stratin g te a ch e r c o m m itm e n t to a ddress b u llie s, b u llie d , and b y stan d e rs.
o & S ' o t.
294
ESPELAGE, POLANIN, AND LOW
w '3 o u O c
o * 3^ < u ,I ;i
^2• c=cs 1 i § £2 § 1 G 33 _ 3 J§ ^ %£ “ >cS #co am >& O Q •2 o . :►h. -g • CO CN m 3 M a * c« 3 3 i— i co
■8 H
i ! C (U c3
,
-S *2 J 5 < u O c33 t3 co O 42 -C -a OX) «(U T3 In h
s
G ■
“ •So* J3 £
,
n G 33
c
G
~
i
s) -§t3 .55~ 3 . «* 3 >r> t o ; co t g ^ s ’ “ j5 j : 3 22 9 % £ £ 1 o 2 | ® o < > S §>w G co . cr < d e'. a c 'O £ ^ en co o 00
3 _j G 2 £ o co _C
_
O
33
«
-3
(U
• G £
£
’
o o Oh a rn co n't
Eigenvalue % Variance
.9 § & « j s a
296
ESPELAGE, POLANIN, AND LOW
School commitment to bully prevention. Eight items emerged in this fourth scale. Teach ers and staff are asked “How much is your school doing in each of the following areas?”: Examples include (1) Demonstrating adminis trator commitment and leadership to address bullies, bullied, and bystanders', (2) Developing policies and programs to prevent bullying', (3) Implementing policies and programs to prevent bullying', and (4) Supporting an active stake holder group to address bullying and guiding implementation o f bullying prevention activi ties. Response options include “Not at all,” “A little bit,” “A fair amount,” and “A lot.” A Cronbach’s a coefficient of .94 was calculated for this sample of teachers and staff. Positive teacher-staff-student interactions. Seven items emerged in the fifth factor. Teach ers and staff are asked how much they agree with statements such as (1) Teachers and staff in this school are willing to help students out', (2) Teachers and staff in this school can be trusted. Response options include “Strongly dis agree,” “Disagree,” “Agree,” and “Strongly agree.” A Cronbach’s a coefficient of .88 was calculated for this sample of teachers and staff. Gender equity or intolerance of sexual harassment. Five items emerged in this sixth factor. Teachers and staff are asked how much they agree with the following statements such as (1) Boys and girls are treated equally in school and (2) Boys and girls show respect fo r each other at school. Response options include “Strongly disagree,” “Disagree,” “Agree,” and “Strongly agree.” A Cronbach’s a coefficient of .79 was calculated for this sample of teachers and staff. School-Level Covariates Percent FRL. The percentage of children who received a FRL was calculated for each school. This is traditionally utilized as a proxy for the school’s average socioeconomic status. Percent female. The percentage of female students was calculated for each school. Percent White. The percentage of White students was calculated for each school. Student Measures Students completed a questionnaire about in volvement in peer aggression, peer victimiza tion, and willingness to intervene as well as a
demographic questionnaire that included ques tions about gender, age, grade, and race or eth nicity. Bullying perpetration. The 9-item Illinois Bully Scale (Espelage & Holt, 2001) assesses the frequency of bullying at school. Students are asked how often in the past 30 days they did the following to other students at school: teased other students, upset other students for the fun of it, excluded others from their group of friends, helped harass other students, and threat ened to hit or hurt another student. Response options include “Never,” “1 or 2 times,” “3 or 4 times,” “5 or 6 times,” and “7 or more times.” Concurrent validity of this scale was established with significant correlations with peer nomina tions of bullying (Espelage, Holt, & Henkel, 2003). Cronbach’s a coefficient was .80. Physical aggression. The 4-item, Univer sity of Illinois Fighting Scale (UIFS; Espelage & Holt, 2001) assesses physical fighting behav ior (e.g., I got in a physical fight; I fought students I could easily beat) the respondent engaged in over the past 30 days. Response options include “Never,” “1 or 2 times,” “3 or 4 times,” “5 or 6 times,” and “7 or more times.” The Fighting Scale had a low corre lation with the Victimization Scale (r = .21) and was only moderately correlated with the Bullying Scale (r = .58), providing evidence of discriminant validity (Espelage & Holt, 2001). Cronbach’s a coefficient was .80. Peer victimization. The 4-item University of Illinois Victimization Scale (Espelage & Holt, 2001) assessed victimization from peers. Students were asked how often the following have happened to them in the past 30 days: Other students called me names, other students made fun o f me, other students picked on me, I got hit and pushed by other students. Response options include “Never,” “1 or 2 times,” “3 or 4 times,” “5 or 6 times,” and “7 or more times.” Cronbach’s a coefficient was .89. W illingness to intervene in bullying episodes. The University of Illinois 5-item Willingness to Intervene in Bully Episodes scale was used to assess students overall will ingness to intervene when other students are being bullied (Espelage, Green, & Polanin, 2012). Students are asked how much they agree with statements about intervening directly or indirectly when they encounter bullying (e.g., “If a kid is being teased, I will stick up for
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
him/her”). Response options include “Strongly disagree,” “Disagree,” “Agree,” and “Strongly agree.” A Cronbach’s a coefficient of .78 was calculated. Analysis Hierarchical linear modeling. The school context is inherently hierarchical because stu dents are nested within schools. Therefore, stu dents’ and teachers’ survey responses remain dependent within schools. Hierarchical linear modeling (HLM) handles this dependency us ing a random effects model to account for cor related errors at the student and school levels (Raudenbush & Bryk, 2002). Before conducting a HLM analysis, however, it is advisable to check the assumptions of re gression and account for missing data (Tabachnick & Fidell, 2007). A histogram and test of skewness indicated a positively skewed distri bution for the four outcomes (2.39, 1.34, 1.71, and .63, respectively, ps < .01), so we deter mined it necessary to conduct sensitivity anal yses. To conduct these analyses, we created a binary outcome and used a logistic regression model; the logit coefficients were not signifi cantly different from the standard multiple re gression model. Therefore, we report only the results from the continuous outcome model. None of the variables, in addition, indicated significant multicollinearity issues. Finally, we accounted for the missing data using multiple imputed datasets (Little & Rubin, 1987). The percentage of missing data was consistent across the four outcome variables at Wave 2 (Perpetration = 26.9, Victimization = 26.5, Physical Aggression = 26.5, Willingness to In tervene = 26.8). Ten imputed datasets were computed using NORM version 2.02 (Schafer, 2002), and following the advice of Enders (2010), we included auxiliary variables in the imputation model as well. Given the prior assumptions met, an HLM analysis was conducted; all model estimations were conducted using HLM 7.1 (Raudenbush, Bryk, & Congdon, 2013). The model process began by estimating an unconditional model with no predictors. This model provided an es timate of the intraclass correlation (ICC; Field, 2009). A model was then fit with Level 1 pre dictors; these coefficients represent similar con structs to traditional regression coefficients. The
297
model-building process continues given statis tically significant random intercepts. Schoollevel variables, Level 2, were then added to model the variation between schools. For our model, we utilized the teacher variables aggre gated to the school level. The Level 1 model can be represented as Yij = Po, + Pi/ * Female + (32/ * Mother’s Education + (33j- * White + (34; * Hispanic + (35y * Asian + (i6/ * Bi-racial + rtj where Yy is one of the three outcomes, (30. is the intercept, ]y is gender difference where female is the reference group, p 2/ is the linear relation ships between the mother’s education and the outcomes, P3j- - (3f)/ is coded to represent the difference in the racial variables where Black was the reference group. Mother’s education was group-mean centered; the other, dichoto mous variables were uncentered. The Level 2 model modeled the school-level variables. The Level 2 model, the same for each outcome, is represented below predicting the school’s intercept: Po, = Too + 7oi * Student Intervention + y 02 * Staff Intervention + y03 * Aggression Problem + y04 * Much School + y05 * Positive Staff —Student Interactions + 7oe * Gender equity/Intolerance of sexual harassment + 707 * FRL+ 708 * % Female + 709 * % White + u0j where 700 represents the grand mean of the outcomes across all schools, and the remainder of the variables represent the relationship be tween the school-level variable and the school’s intercept. All variables at Level 2 were grand-
ESPELAGE, POLANIN, AND LOW
298
mean centered. To evaluate the variables and the model, traditional statistical significance testing was conducted; we also calculated the standardized regression coefficients to allow for coefficient comparison (Tabachnick & Fidell, 2007). We also calculated the percentage of Level 2 variance accounted for by the model after three model-building stages (Table 5; Raudenbush & Bryk, 2002). These calculations allowed us to estimate the variables’ contribu tions to the model.
Results School-Level Associations Between Teacher and Staff Perceptions and Student Self-Report Given the context of the teacher and staff factors, we aggregated the variables to the school level and estimated the intercorrelations (see Table 3). As hy pothesized, greater perceptions of staff intervention to stop bullying were significantly associated with higher student self-reported willingness to intervene (,r = .41), but not associated with lower bullying, aggression, or victimization. Student intervention re ported by the teachers and staff was not associated with any of the student reports, which did not support our hypothesis. As hypothesized, in schools where teachers and staff viewed aggression as a problem, students reported significantly greater bullying (r = .55), aggression (r = .52), victimization (r = .63), and less willingness to intervene (r = —.61). Further, when school staff perceived administrative support for bullying or violence prevention, students reported significantly less bullying (r = —.34), aggression (r = —.35), victimization (r = —.31), and greater willingness to intervene (r = .38). Support was also found for our hypotheses that positive teacherstudent relations and gender equity or intolerance of sexual harassment at the school level would be sig nificantly associated with less bullying (rs = —.54, —.52), aggression (rs = —.46, —.56), victimization (rs = —.55, —.59), and greater willingness to inter vene (rs = .53, .60).
Hierarchical Linear Modeling An unconditional model, where no Level 1 or Level 2 predictors were included, allowed for the estimated of the ICCs. The results of this model indicated ICCs of .061, .04, .093, and .041 for the four outcomes: bullying perpetra
tion, bullying victimization, physical aggres sion, and willingness to intervene, respectively. Therefore, we allowed the intercept to vary ran domly and included Level 1 and Level 2 vari ables into the model (see Table 4). Given that the focus of this project was on the Level 2 variables, the discussion of the significant re sults is constrained to these variables only. Bullying perpetration. Teacher-perceptions of administrators support around addressing bullying and violence was significantly related to lower levels of student self-reported bullying perpetration (702 = —.20, SE = .06, p < .01, B = -.1 3 ). Greater teacher perceptions of gen der equity or intolerance of sexual harassment at the school-level was significantly associated with less student self-reported bullying perpe tration (y02 = - .2 3 , SE = .10, p < .05, B = —.08). Schools with a greater percentage of females had significantly less bullying perpetra tion (-y02 = —.66, SE = .29, p < .05, B = —.07). The Level 2 variables accounted for 71.4% of the variance at the school level; 28.65% of the variance accounted for was because of student characteristics and dem o graphic school characteristics. The remaining 42.8% of the variance was accounted for by the teacher and staff variables. Peer victimization. The results revealed several school-level variables that predicted bullying victim ization at the student level. Teachers and staff perceptions that their ad m inistration was supportive of them address ing bullying and violence was associated with less self-reported peer victim ization by stu dents (704 = —.42, SE = .09, p < .01, B = - .2 7 ) . Schools that had a high level of gender equity or intolerance of sexual harassment also tended to have lower peer victimization (y 04 = —.71, SE = .20, p < .01, B = - .2 4 ). Schools with a greater percentage of W hite students (y 09 = —.49, SE = .18,/? < .05, B = —.18) and female students (708 = —.71, SE = .32, p < .05, B = —.08) also indicated sig nificantly less student-reported peer victim ization as well. The Level 2 variables ac counted for 77.5% of the variance; 74.3% of the variance was accounted for by the teacher and staff school-level variables. Physical aggression. School-level analy ses indicated that the schools specifically ad dressing bullying and violence have signifi cantly less physical aggression (7 04 = —.17,
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
299
I I
“
o f
I
I I
I I
I I I
I I I
I I
I I
o f
8 2 I I
o f
S
3 U~) 3 «n CO
— CO CN
o p CO
O
w
O
r- N o < CO
§
D
— -c c U +J o cd G
c D 4) 2 £ ! 'oo C 36 dg- ~> -o 2fl ' C ' U •8 1 3 u S '1 1 h 2 1■ rt
u
3
.2
.
s ; U &■ ^ • O X
—
p
ON p NO C CN
»o
i s B$ £ £
IS 5 g3 m g '— ap p U
School Psychology Quarterly 2014, Vol. 29, No. 3, 287-305
Teacher and Staff Perceptions of School Environment as Predictors of Student Aggression, Victimization, and Willingness to Intervene in Bullying Situations Dorothy L. Espelage
Joshua R. Polanin
University of Illinois at Champaign-Urbana
Vanderbilt University
Sabina K. Low Arizona State University This study examines how teacher and staff perceptions of the school environment correlate with student self-reports of bullying, aggression, victimization, and willing ness to intervene in bullying incidents using multi-informant, multilevel modeling. Data were derived from 3,616 6th grade students across 36 middle schools in the Midwest, who completed survey measures of bullying, aggression, victimization, and willingness to intervene in bullying situations. Teachers and staff (n = 1,447) com pleted a school environment survey. Bivariate associations between school-level and student self-reports indicated that as teacher and staff perceive aggression as a problem in their school, students reported greater bully perpetration, fighting, peer victimization, and less willingness to intervene. Further, as staff and teacher report greater commit ment to prevent bullying and viewed positive teacher and student relationships, there was less bullying, fighting, and peer victimization, and greater willingness to intervene. In a model where all school environment scales were entered together, a school commitment to prevent bullying was associated with less bullying, fighting, and peer victimization. Student-reports of bully perpetration and peer victimization were largely explained by staff and teacher commitment to bully prevention, whereas fighting and willingness to intervene were largely explained by student characteristics (e.g., gender). We conclude that efforts to address bullying and victimization should involve support from the school administration. School psychologists should play an active role in the school climate improvement process, by creating a school climate council consisting of students, parents, and teachers; administering school climate measures; identifying specific school improvement targets from these data, and engaging all stakeholders in the ongoing school improvement plan. Keywords: bullying, school climate, multilevel modeling, middle school, teacher
Com prehensive school-w ide efforts to pre vent aggression, bullying, and victim ization are predicated on the assum ption that reducing these behaviors m ust include w orking w ith all stakeholders in the school com m unity, includ
ing teachers, students, and parents (Espelage, in press; O lw eus & L im ber, 2010). D raw ing from the ecological m odel o f child developm ent, the school environm ent is an im portant m icrosys tem that influences how students’ view bullying
This article was published Online First August 4, 2014. Dorothy L. Espelage, Department of Educational Psy chology, University of Illinois at Urbana-Champaign; Joshua R. Polanin, Peabody Research Institute, Vanderbilt University; Sabina K. Low, School of Social and Family Dynamics, Arizona State University. Research for the current study was supported by the Centers for Disease Control & Prevention (1U01/ CE001677) to Dorothy L. Espelage (PI) at the Univer
sity of Illinois at Urbana-Champaign. Opinions ex pressed herein do not necessarily reflect those of the Centers for Disease Control & Prevention, or related offices within. Correspondence concerning this article should be ad dressed to Dorothy L. Espelage, Department of Educa tional Psychology, University of Illinois at UrbanaChampaign, Champaign, IL 61820-6925. E-mail: espelage @illinois.edu 287
288
ESPELAGE, POLANIN, AND LOW
and aggression, how adult role models correlate with student behavior, and how school-level norms and policies correlate with these student behaviors (Bronfenbrenner, 1977; Espelage, in press). Studies consistently report that negative school environmental factors (e.g., policies, staff reaction to bullying) can lead to an in crease in the frequency of bullying, aggression, and victimization and reduce the likelihood of students feeling safe in their school (Espelage, Bosworth, & Simon, 2000; Goldweber, Waasdorp, & Bradshaw, 2013). In contrast, youth with positive perceptions of their school envi ronment are less likely to exhibit externalizing behaviors (e.g., aggression; Espelage et al., 2000; Goldweber et al., 2013; Totura et al., 2009). The majority of these studies, however, are limited to student self-report of school cli mate or a narrow assessment of school environ ment (e.g., safety, belonging). The current study addresses this gap in the literature by examining teacher and staff perceptions of the school en vironment and how these perceptions correlate with student self-reports of bullying, aggres sion, victimization, and willingness to intervene in bullying incidents. To examine the mac rolevel correlations with student behavior that move beyond the classroom level, we utilize a broader, multidimensional assessment of school environment to include a comprehensive focus and commitment to bully prevention and sup ports to teacher and staff to learn ways to stop bullying. In addition, we use multilevel model ing to capture precisely student behaviors as a function of the larger social context at the school-level. School Environment and Climate Predictors School environment is a broad term that en compasses multiple features of school climate or “culture,” and in this article, refers to the psychosocial quality and character of school life (i.e., we did not assess physical qualities of the school; G ottfredson, Gottfredson, Payne, & Gottfredson, 2005). School climate is based on patterns of people’s experience of school life and reflects norms, goals, values, interpersonal relationships, teaching, learn ing, leadership practices, and organizational structures (National School Climate Council, 2007). Because of frequent interactions be
tween teachers and students in school, it is not surprising that teacher’s and other staff’s per ceptions of the environment are associated with student attitudes and behaviors. In a study of 40 countries, Harel-Fisch and col leagues (2011) analyzed the World Health Organization health behavior in school-age children (WHO-HBSC) surveys and found that as negative school perceptions reported by students increased, so did their involve ment in bullying as a perpetrator or victim. Building on these findings, in the current study we examine how six indices of school climate and environment from the teacherstaff perspective correlate with bullying, vic timization, aggression, and willingness to in tervene in bullying situations from student report. The school climate and environment scales includes perceptions of (a) staff and student intervention in bullying, (b) aggres sion as a problem, (c) administrative support and opportunities to prevent bullying and vi olence, (d) positive teacher-staff-student rela tions, and (e) gender equity and intolerance of sexual harassment. Staff and Student Intervention Research indicates that bully perpetration and victimization rates are higher and willingness to intervene is lower when students perceive adults’ prevention and intervention efforts as ineffective (Goldweber et al., 2013; Waasdorp, Pas, O’Breenan, & Bradshaw, 2011). Therefore, we hypothesized that greater levels of teacher and staff intervention would be associated with less student reports of bullying, victimization, aggression, and greater willingness to intervene. Aggression as a Problem It has been noted that there are discrepancies between how teachers and staff perceive bully ing in comparison to their students. Many teach ers are unaware of how serious and extensive the bullying is within their schools, and are often ineffective in being able to identify bul lying incidents (Bradshaw, Sawyer, & O’Brennan, 2007; Waasdorp et al., 2011). Di vergence between staff and student estimates of the rates of bullying are seen in elementary, middle, and high school, with staff consistently underestimating the frequency of these events (Bradshaw et al., 2007). In a large-scale valida-
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
tion study of three climate measures, Bandy opadhyay and colleagues (2009) found signifi cant, albeit modest, correlations between teacher and student reports of bullying and teas ing across 291 Virginia high schools. Thus, we hypothesize that teacher and staff perceptions of aggression at the school-level will be signifi cantly associated with student reports of bully ing, fighting, and victimization.
sexual harassment or treat boys and girls differ ently, students report greater peer victimization and less willingness to seek help for themselves or other peers (Charmaraman, Jones, Stein, & Espelage, 2013). Therefore, it was hypothesized that lower bullying and victimization rates and greater willingness to intervene would be asso ciated with greater gender equity or intolerance of sexual harassment at the teacher-staff level.
School Commitment to Bully Prevention Orpinas and Home (2006) argue that bully occurs less often in schools where the leader ship provided by the school administration works closely with staff to understand policies and seek input from all adults. Twemlow and Sacco (2013) argue that bullying is a universal dysfunctional social process, where bullying and victimization at the student level are symp toms of an adult culture that are not attending to the issue at hand. When adults deny the prob lem, youth report greater aggression and victim ization. Thus, we hypothesized that selfreported student levels of bullying, aggression, and victimization would be lower in schools where teachers and staff perceive that prevent ing bullying and violence is a priority and they are provided with professional development op portunities. Positive Teacher-Staff-Student Interactions Youth with lower levels of school connect edness were significantly more likely to be in volved in bullying and peer victimization (Espelage et al., 2000; Glew et al., 2005; Goldweber et al., 2013). Thus, we hypothesized that closer teacher-student relations reported by the teachers and staff would be associated with less bullying, aggression, victimization, and greater willingness to intervene by students.
289
Current Study In summary, the extant research literature suggests that the school environment is associ ated with student’s self-reported bullying, ag gression, victimization, and student’s willing ness to intervene to help other students. However, only a few studies have used multi level analyses to fully capture the associations of school climate on student attitudes and be haviors, but these studies are limited. Waasdorp and colleagues (2011), for example, used mul tilevel modeling with student, staff, and parent data, but used single items to assess school safety, belonging, witnessing bullying, and bully perpetration and victimization. Richard and colleagues (2012) surveyed over 18,000 French youth and examined school-level char acteristics using multilevel modeling largely from aggregated student reports because only 701 teachers completed measures. This current study adds to the growing body of literature on bullying and school climate by assessing per ceptions of school environment from teachers and staff across multiple scales and selfreported behaviors from students. Furthermore, the teacher and staff environment surveys were completed by teachers, paraprofessionals, ad ministrators, and other adults in each school, offering a broader assessment of the school environment.
Gender Equity and Intolerance of Sexual Harassment
Method Participants
Verbal aggression, peer victimization, or bul lying during early adolescence can involve ho mophobic name-calling and sexual commentary or sexual touching, and when these are un wanted they are referred to as sexual harassment or sexual violence (Espelage, Basile, & Ham burger, 2012; Poteat & Espelage, 2007). When the adults in middle school are dismissive of
In total, 3,616 6th grade students across Illi nois and Kansas middle schools participated in the surveys (see Table 1). The students ranged in age from 11 to 13 years old, but most stu dents were 11 (Illinois = 75.8%, Kansas = 78.4%). A slight majority of students were male in both Illinois and Kansas (Illinois = 51.2%,
290
ESPELAGE, POLANIN, AND LOW
Table 1 Participant and School Characteristics Illinois Number of students Number of schools Age 11 12 13 Gender Male Female Race Black Asian Latina/o White Biracial Mother’s education Less than high school High school graduate Some college Graduate from college Some or all postgraduate Unknown/no response Father’s education Less than high school High school graduate Some college Graduate from college Some or all postgraduate Unknown/no response Teacher and staff N Average N per school Median N per school Percent free or reduce lunch Percent females Percent White Percent Black
2,012 24
Kansas 1,604 12
75.8 22.5 1.7
78.4 20.7 0.9
51.2 48.8
52.9 47.1
33.5 1.0 33.1 20.2 11.7
17.5 4.2 35.2 30.4 12.8
17.9 25.8 19.0 20.2 9.9 7.2
13.3 19.0 15.4 15.7 7.3 29.2
19.2 25.9 15.4 17.1 9.8 13.6 732 30.08 27.50 71.95 49.4 19.4 43.5
13.8 19.3 10.6 13.0 6.0 37.3 715 59.08 58.50 78.52 42.8 33.0 19.1
Kansas = 52.9%). In Illinois, Blacks (33.5%) and Latino/as (33.1%) constituted the majority of students surveyed. Most students in Kansas schools were Latino/a (35.2%) or White (30. 4%). Overall, 73% of students were eligible for free or reduced lunch (FRL) across both states. States did not differ significantly with regard to the percentage of FRL children or percentage of females (Illinois = 71.9, Kansas = 78.9). Illi nois students indicated that the majority of their mothers graduated from college (20.2%) com pared with mothers of Kansas students (15.7%). A large portion of students from Kansas, how ever, either did not know their mother’s educa tion or left this item blank.
The analyses presented here include baseline data from a large-scale randomized clinical trial of a social-emotional learning program that is described in more detail elsewhere (Espelage, Low, Polanin, & Brown, 2013). The sample, therefore, was drawn initially to test the efficacy of this program in a nested cohort (6th graders) longitudinal study. Schools were recruited through the school district offices and had to be willing to be assigned to either an intervention or control condition. To be eligible, the schools could not be implementing any large-scale bully prevention curriculum or initiative, and control schools agreed to not implement a bully preven tion program until after the 3 year trial. Twentyfour schools from Illinois and 12 schools from Kansas participated in the project. All staff were invited to complete the survey, not just 6th grade teachers. In Illinois schools, 732 teachers and other staff completed the survey and 715 teachers and other staff from Kansas schools participated. One school in Illinois provided only two teacher and staff surveys; all other schools, across both states, had at least 10 teach ers participate in the survey (range = 2-101; M = 39.69; SD = 24.24). Schools were eligible for the school-level stipend if 80% of their teachers and staff completed the survey; all but three schools reached this goal. The sample consisted of 66% teachers, 10% support staff, 9% paraprofessionals, 4% administrators, 3% counselors or psychologists, 2% custodian staff, and 1% cafeteria staff. The staff identified mainly as White (75%), 10% as Black, and 8% as Hispanic, and 78% of the sample identified as female. The average age of the sample was 42.5 years old. Data Collection A waiver of active parental consent and an active consent protocol were both approved by the institutional review board at the University of Illinois, and districts could use either method. Parents of all 6th grade students enrolled in all participating schools were sent letters or con sent forms. An 86% participation rate was achieved in schools using a waiver of active consent and 63% participation rate was achieved for schools using an active consent procedure. Students were asked to consent to participate in the study through an assent pro cedure included on the coversheet of the survey.
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
At each wave of data collection, six trained research assistants, the primary researcher, and a faculty member collected the data. At least two of these individuals administered surveys to classes ranging in size from 10 to 25 students. The research assistants first informed students about the general nature of the investigation. Students were then given survey packets and the survey was read aloud to them. It took students ~ 40 min to complete the survey. Teachers and staff at each school were sent an email with a link to the online School Environment Survey. Teacher and Staff School Environment Survey Data were collected from school staff mem bers using the School Environment Survey. This instrument was adapted from the Colorado Trust’s Bullying Prevention Initiative (Csuti, 2008) for an elementary school study. Demo graphic information related to staff members’ gender, race or ethnicity, age, position in the school, and length of employment at the school is included in the survey. Because the current study was conducted in middle schools and this measure was first validated with elementary schools, these items were subjected to factor analysis to determine the scale score. To deter mine the latent factor structure of the school environment survey variables, an exploratory factor analysis (EFA) was utilized (Kahn, 2006). The EFA followed the procedures out lined in Kline (2011). Concern was given to the number of response categories utilized, but Kline (2011) indicated that four responses were reasonable for maximum-likelihood estimation procedures. A principal axis factoring was uti lized to extract the factors and a varimax rota tion was implemented to rotate the structure. A scree plot and subsequent parallel analysis (Hayton, Allen, & Scarpello, 2004) confirmed the latent factor structure and the final model was fixed to extract the exact number of factors required (8). Items that loaded above .3 on the hypothesized factor were retained (Kline, 2011); items that failed to meet these criteria were subsequently dropped from the variable construction. We did not consider the clustered nature of the data for the factor analysis because the statistical significance of the factor loadings was not considered as important as the factor loadings and because the latent factors were
291
aggregated to the school-level (Stapleton, 2006). We conducted the factor analysis in SPSS version 21 (IBM Corp., 2013); we uti lized O’connor’s (2000) SPSS macro to conduct the parallel analysis. Results of the principal axis factor explor atory factor analysis revealed nine potential fac tors. Upon further review of the scree plot and parallel analysis, however, only eight factors were perceived as viable. Six of the eight fac tors identified were used in this manuscript. The two other factors were considered not theoreti cally important and, therefore, dropped from the analyses. The six factors’ item loadings, eigen values, percentage of variance, and Cronbach’s a coefficients are detailed in Table 2. The Cronbach's a coefficients ranged from .79 to .94. Overall, the six factors accounted for 56.8% of the total variance. Student intervention. Five items emerged in this first factor. Teachers and staff are asked “How likely is it that STUDENTS at your school could be counted on to help out in the following situations?” Examples include (1) A student is making fun o f and teasing another student who is obviously weaker, (2) A student is spreading rumors or lies about another stu dent behind their back. Response options in clude “Very unlikely,” “Unlikely,” “Likely,” and “Very likely.” A Cronbach’s a coefficient of .84 was calculated for this sample of teachers and staff. Staff intervention. Five items emerged in this second factor. Teachers and staff are asked “How likely is it that STAFF at your school could be counted on to help out in the following situations?” And the same five items from the student intervention (above) are presented. Re sponse options include “Very u n l i k e l y “Un likely,” “Likely,” and “Very likely.” A Cron bach’s a coefficient of .91 was calculated for this sample of teachers and staff. Perceptions of aggression as a problem. Five items emerged in this third factor. Teach ers and staff are asked “How much of a problem are the following issues at your school?” Exam ples include (1) Students picking fights with other students', (2) Students who push, shove, or trip weaker students. Response options include “Not a problem,” “Sort o f a problem,” “Pretty big problem,” and “Huge problem.” A Cron bach’s a coefficient of .87 was calculated for this sample of teachers and staff.
ESPELAGE, POLANIN, AND LOW
292
CT O d c 2 cn R
Standardized Factor Loadings, Eigen Values, % Variance, and Cronbach’s a Coefficients for Teacher and Staff Variables
Table 2
o •
c
c
3 o .5 CQ " s I o
3 2 X 0 00 0 X £ 33 0c C/5 rQ ^ 03 c = x: y .3 -e o .s § M 03 :■ s-o ,c :5 *3 • — 03 o 3 . -P M -a - ^ o> a2 -c : 3 -o 00 "2 3 < £ -c 3 o c/3 3 3 3 o c 35 ^ . C a/5- SO £ o c/5 T3 . C C/5 i & cd . C/5 CO >o o
•a
17. D e m o n stratin g te a ch e r c o m m itm e n t to a ddress b u llie s, b u llie d , and b y stan d e rs.
o & S ' o t.
294
ESPELAGE, POLANIN, AND LOW
w '3 o u O c
o * 3^ < u ,I ;i
^2• c=cs 1 i § £2 § 1 G 33 _ 3 J§ ^ %£ “ >cS #co am >& O Q •2 o . :►h. -g • CO CN m 3 M a * c« 3 3 i— i co
■8 H
i ! C (U c3
,
-S *2 J 5 < u O c33 t3 co O 42 -C -a OX) «(U T3 In h
s
G ■
“ •So* J3 £
,
n G 33
c
G
~
i
s) -§t3 .55~ 3 . «* 3 >r> t o ; co t g ^ s ’ “ j5 j : 3 22 9 % £ £ 1 o 2 | ® o < > S §>w G co . cr < d e'. a c 'O £ ^ en co o 00
3 _j G 2 £ o co _C
_
O
33
«
-3
(U
• G £
£
’
o o Oh a rn co n't
Eigenvalue % Variance
.9 § & « j s a
296
ESPELAGE, POLANIN, AND LOW
School commitment to bully prevention. Eight items emerged in this fourth scale. Teach ers and staff are asked “How much is your school doing in each of the following areas?”: Examples include (1) Demonstrating adminis trator commitment and leadership to address bullies, bullied, and bystanders', (2) Developing policies and programs to prevent bullying', (3) Implementing policies and programs to prevent bullying', and (4) Supporting an active stake holder group to address bullying and guiding implementation o f bullying prevention activi ties. Response options include “Not at all,” “A little bit,” “A fair amount,” and “A lot.” A Cronbach’s a coefficient of .94 was calculated for this sample of teachers and staff. Positive teacher-staff-student interactions. Seven items emerged in the fifth factor. Teach ers and staff are asked how much they agree with statements such as (1) Teachers and staff in this school are willing to help students out', (2) Teachers and staff in this school can be trusted. Response options include “Strongly dis agree,” “Disagree,” “Agree,” and “Strongly agree.” A Cronbach’s a coefficient of .88 was calculated for this sample of teachers and staff. Gender equity or intolerance of sexual harassment. Five items emerged in this sixth factor. Teachers and staff are asked how much they agree with the following statements such as (1) Boys and girls are treated equally in school and (2) Boys and girls show respect fo r each other at school. Response options include “Strongly disagree,” “Disagree,” “Agree,” and “Strongly agree.” A Cronbach’s a coefficient of .79 was calculated for this sample of teachers and staff. School-Level Covariates Percent FRL. The percentage of children who received a FRL was calculated for each school. This is traditionally utilized as a proxy for the school’s average socioeconomic status. Percent female. The percentage of female students was calculated for each school. Percent White. The percentage of White students was calculated for each school. Student Measures Students completed a questionnaire about in volvement in peer aggression, peer victimiza tion, and willingness to intervene as well as a
demographic questionnaire that included ques tions about gender, age, grade, and race or eth nicity. Bullying perpetration. The 9-item Illinois Bully Scale (Espelage & Holt, 2001) assesses the frequency of bullying at school. Students are asked how often in the past 30 days they did the following to other students at school: teased other students, upset other students for the fun of it, excluded others from their group of friends, helped harass other students, and threat ened to hit or hurt another student. Response options include “Never,” “1 or 2 times,” “3 or 4 times,” “5 or 6 times,” and “7 or more times.” Concurrent validity of this scale was established with significant correlations with peer nomina tions of bullying (Espelage, Holt, & Henkel, 2003). Cronbach’s a coefficient was .80. Physical aggression. The 4-item, Univer sity of Illinois Fighting Scale (UIFS; Espelage & Holt, 2001) assesses physical fighting behav ior (e.g., I got in a physical fight; I fought students I could easily beat) the respondent engaged in over the past 30 days. Response options include “Never,” “1 or 2 times,” “3 or 4 times,” “5 or 6 times,” and “7 or more times.” The Fighting Scale had a low corre lation with the Victimization Scale (r = .21) and was only moderately correlated with the Bullying Scale (r = .58), providing evidence of discriminant validity (Espelage & Holt, 2001). Cronbach’s a coefficient was .80. Peer victimization. The 4-item University of Illinois Victimization Scale (Espelage & Holt, 2001) assessed victimization from peers. Students were asked how often the following have happened to them in the past 30 days: Other students called me names, other students made fun o f me, other students picked on me, I got hit and pushed by other students. Response options include “Never,” “1 or 2 times,” “3 or 4 times,” “5 or 6 times,” and “7 or more times.” Cronbach’s a coefficient was .89. W illingness to intervene in bullying episodes. The University of Illinois 5-item Willingness to Intervene in Bully Episodes scale was used to assess students overall will ingness to intervene when other students are being bullied (Espelage, Green, & Polanin, 2012). Students are asked how much they agree with statements about intervening directly or indirectly when they encounter bullying (e.g., “If a kid is being teased, I will stick up for
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
him/her”). Response options include “Strongly disagree,” “Disagree,” “Agree,” and “Strongly agree.” A Cronbach’s a coefficient of .78 was calculated. Analysis Hierarchical linear modeling. The school context is inherently hierarchical because stu dents are nested within schools. Therefore, stu dents’ and teachers’ survey responses remain dependent within schools. Hierarchical linear modeling (HLM) handles this dependency us ing a random effects model to account for cor related errors at the student and school levels (Raudenbush & Bryk, 2002). Before conducting a HLM analysis, however, it is advisable to check the assumptions of re gression and account for missing data (Tabachnick & Fidell, 2007). A histogram and test of skewness indicated a positively skewed distri bution for the four outcomes (2.39, 1.34, 1.71, and .63, respectively, ps < .01), so we deter mined it necessary to conduct sensitivity anal yses. To conduct these analyses, we created a binary outcome and used a logistic regression model; the logit coefficients were not signifi cantly different from the standard multiple re gression model. Therefore, we report only the results from the continuous outcome model. None of the variables, in addition, indicated significant multicollinearity issues. Finally, we accounted for the missing data using multiple imputed datasets (Little & Rubin, 1987). The percentage of missing data was consistent across the four outcome variables at Wave 2 (Perpetration = 26.9, Victimization = 26.5, Physical Aggression = 26.5, Willingness to In tervene = 26.8). Ten imputed datasets were computed using NORM version 2.02 (Schafer, 2002), and following the advice of Enders (2010), we included auxiliary variables in the imputation model as well. Given the prior assumptions met, an HLM analysis was conducted; all model estimations were conducted using HLM 7.1 (Raudenbush, Bryk, & Congdon, 2013). The model process began by estimating an unconditional model with no predictors. This model provided an es timate of the intraclass correlation (ICC; Field, 2009). A model was then fit with Level 1 pre dictors; these coefficients represent similar con structs to traditional regression coefficients. The
297
model-building process continues given statis tically significant random intercepts. Schoollevel variables, Level 2, were then added to model the variation between schools. For our model, we utilized the teacher variables aggre gated to the school level. The Level 1 model can be represented as Yij = Po, + Pi/ * Female + (32/ * Mother’s Education + (33j- * White + (34; * Hispanic + (35y * Asian + (i6/ * Bi-racial + rtj where Yy is one of the three outcomes, (30. is the intercept, ]y is gender difference where female is the reference group, p 2/ is the linear relation ships between the mother’s education and the outcomes, P3j- - (3f)/ is coded to represent the difference in the racial variables where Black was the reference group. Mother’s education was group-mean centered; the other, dichoto mous variables were uncentered. The Level 2 model modeled the school-level variables. The Level 2 model, the same for each outcome, is represented below predicting the school’s intercept: Po, = Too + 7oi * Student Intervention + y 02 * Staff Intervention + y03 * Aggression Problem + y04 * Much School + y05 * Positive Staff —Student Interactions + 7oe * Gender equity/Intolerance of sexual harassment + 707 * FRL+ 708 * % Female + 709 * % White + u0j where 700 represents the grand mean of the outcomes across all schools, and the remainder of the variables represent the relationship be tween the school-level variable and the school’s intercept. All variables at Level 2 were grand-
ESPELAGE, POLANIN, AND LOW
298
mean centered. To evaluate the variables and the model, traditional statistical significance testing was conducted; we also calculated the standardized regression coefficients to allow for coefficient comparison (Tabachnick & Fidell, 2007). We also calculated the percentage of Level 2 variance accounted for by the model after three model-building stages (Table 5; Raudenbush & Bryk, 2002). These calculations allowed us to estimate the variables’ contribu tions to the model.
Results School-Level Associations Between Teacher and Staff Perceptions and Student Self-Report Given the context of the teacher and staff factors, we aggregated the variables to the school level and estimated the intercorrelations (see Table 3). As hy pothesized, greater perceptions of staff intervention to stop bullying were significantly associated with higher student self-reported willingness to intervene (,r = .41), but not associated with lower bullying, aggression, or victimization. Student intervention re ported by the teachers and staff was not associated with any of the student reports, which did not support our hypothesis. As hypothesized, in schools where teachers and staff viewed aggression as a problem, students reported significantly greater bullying (r = .55), aggression (r = .52), victimization (r = .63), and less willingness to intervene (r = —.61). Further, when school staff perceived administrative support for bullying or violence prevention, students reported significantly less bullying (r = —.34), aggression (r = —.35), victimization (r = —.31), and greater willingness to intervene (r = .38). Support was also found for our hypotheses that positive teacherstudent relations and gender equity or intolerance of sexual harassment at the school level would be sig nificantly associated with less bullying (rs = —.54, —.52), aggression (rs = —.46, —.56), victimization (rs = —.55, —.59), and greater willingness to inter vene (rs = .53, .60).
Hierarchical Linear Modeling An unconditional model, where no Level 1 or Level 2 predictors were included, allowed for the estimated of the ICCs. The results of this model indicated ICCs of .061, .04, .093, and .041 for the four outcomes: bullying perpetra
tion, bullying victimization, physical aggres sion, and willingness to intervene, respectively. Therefore, we allowed the intercept to vary ran domly and included Level 1 and Level 2 vari ables into the model (see Table 4). Given that the focus of this project was on the Level 2 variables, the discussion of the significant re sults is constrained to these variables only. Bullying perpetration. Teacher-perceptions of administrators support around addressing bullying and violence was significantly related to lower levels of student self-reported bullying perpetration (702 = —.20, SE = .06, p < .01, B = -.1 3 ). Greater teacher perceptions of gen der equity or intolerance of sexual harassment at the school-level was significantly associated with less student self-reported bullying perpe tration (y02 = - .2 3 , SE = .10, p < .05, B = —.08). Schools with a greater percentage of females had significantly less bullying perpetra tion (-y02 = —.66, SE = .29, p < .05, B = —.07). The Level 2 variables accounted for 71.4% of the variance at the school level; 28.65% of the variance accounted for was because of student characteristics and dem o graphic school characteristics. The remaining 42.8% of the variance was accounted for by the teacher and staff variables. Peer victimization. The results revealed several school-level variables that predicted bullying victim ization at the student level. Teachers and staff perceptions that their ad m inistration was supportive of them address ing bullying and violence was associated with less self-reported peer victim ization by stu dents (704 = —.42, SE = .09, p < .01, B = - .2 7 ) . Schools that had a high level of gender equity or intolerance of sexual harassment also tended to have lower peer victimization (y 04 = —.71, SE = .20, p < .01, B = - .2 4 ). Schools with a greater percentage of W hite students (y 09 = —.49, SE = .18,/? < .05, B = —.18) and female students (708 = —.71, SE = .32, p < .05, B = —.08) also indicated sig nificantly less student-reported peer victim ization as well. The Level 2 variables ac counted for 77.5% of the variance; 74.3% of the variance was accounted for by the teacher and staff school-level variables. Physical aggression. School-level analy ses indicated that the schools specifically ad dressing bullying and violence have signifi cantly less physical aggression (7 04 = —.17,
TEACHER AND STAFF PERCEPTIONS OF AGGRESSION
299
I I
“
o f
I
I I
I I
I I I
I I I
I I
I I
o f
8 2 I I
o f
S
3 U~) 3 «n CO
— CO CN
o p CO
O
w
O
r- N o < CO
§
D
— -c c U +J o cd G
c D 4) 2 £ ! 'oo C 36 dg- ~> -o 2fl ' C ' U •8 1 3 u S '1 1 h 2 1■ rt
u
3
.2
.
s ; U &■ ^ • O X
—
p
ON p NO C CN
»o
i s B$ £ £
IS 5 g3 m g '— ap p U
