Examining the Differential Relationships Between School Climate and Students’ Disciplinary Outcomes: Evidence from New York City

Abstract

Improving school climate has emerged as a strategy to reduce discipline disparities. This study uses the case of New York City to examine the differential relationships between school climate and students’ likelihood of receiving office discipline referrals (ODRs) or suspensions. The results illustrate that the association between school climate and school discipline varies significantly by student characteristics. Black, Latinx, male, low-income, special education status and students in temporary housing have a smaller associated decrease in likelihood of receiving ODRs and suspensions in more positive school climates relative to other students, although this depends on whether disparities metrics are absolute or relative. School racial composition, school achievement, and the diversity of teachers and school leaders moderate the association between school climate and disciplinary outcomes for Black and Latinx students. Policymakers should consider concurrent investments in school climate and the diversity of school personnel that influences who benefits most from school climate.

Persistent racial disparities in exclusionary discipline are a pressing concern among K-12 education stakeholders (Welsh, 2023; Welsh & Shafiqua, 2018a). Discipline disparities have persisted in the post-pandemic era. The latest data from the Civil Rights Data Collection (2021–2022) indicate that Black boys were 8% of the K-12 student enrollment but 18% of students who received one or more ISS and 22% of students who received one or more OSS, and Black girls were 7% of K-12 student enrollment but 11% of students who received one or more ISS and 13% of students who received one or more OSS (Civil Rights Data Collection, 2025). Black students lose the most instructional time due to suspensions, and this lost time is linked to lower achievement and adult crime (Bacher-Hicks et al., 2019; Darling-Hammond & Ho, 2024; Losen & Whitaker, 2018; Welsh & Shafiqua, 2018b). Owens and McLanahan (2020) examined the drivers of racial disparities in suspensions and expulsions and found that the differential treatment and support of students with similar behaviors accounted for the majority of disparities in exclusionary discipline. As argued by Milner (2020), contemporary school discipline can be characterized as punishment for Black students whose behavior is misaligned with White educators’ behavioral and academic expectations.

Reducing racial inequality in exclusionary discipline is a pivotal component of improving K-12 education systems. Despite myriad efforts at the federal, state and district levels, discipline disparities seem immune to school discipline reforms (Welsh, 2023; Welsh & Shafiqua, 2018a; Welsh et al., 2023). School climate—a manipulatable factor influenced by educational policymaking at the school, district, state and federal levels—is the factor of interest in this study. Scholars and the federal government have posited improving school climate as a way to reduce racial inequality in school discipline (Bradshaw et al., 2009; Gage et al., 2016; Huang & Cornell, 2018; Mitchell et al., 2010; Skiba et al., 2014; U.S. Department of Justice and U.S. Department of Education, 2014; Welsh et al., 2024a). Popular school-based programs such as Restorative Justice (RJ) are intended to reduce the use of and racial disparities in suspensions through improvements in school climate or the quality and character of school life (Gregory & Evans, 2020; Gregory et al., 2018; Payne & Welch, 2022). Indeed, the relationship between school discipline and school climate is a complex one and may partly answer the bigger question: Why have school discipline reforms reduced rates but not narrowed disparities in students’ disciplinary outcomes? A key element of the answer to this question is the possibility of differential relationships between school climate and students’ disciplinary outcomes.

Prior studies found that school climate moderates the relationships between student socioeconomic background and achievement (Berkowitz et al., 2017). It is reasonable to expect that school climate may not benefit all students equally (Wang & Degol, 2016). Yet whether school climate has differential benefits for students of color has been largely overlooked (Wang & Degol, 2016). Wang and Degol (2016) highlighted that “most studies to date have relied upon variable-centered approaches instead of person-centered approaches to study school climate. . . Given the lack of person-centered approaches in school climate research, the examination of heterogeneity and varying configurations among domains of school climate has been vastly overlooked” (p. 338). Few studies have explored the differential associations of school climate with a student’s likelihood of receiving an office discipline referral (ODR) or suspension (Gregory et al., 2018; Heilbrun et al., 2018; Huang & Cornell, 2018). The limited empirical evidence on whether Black, Latinx, and low-income students benefit differentially from school climate improvements has found mixed results. In addition, recent studies have highlighted the distinctions between school climate and racial climate as well as the importance of considering race and intersectionality to understanding school climate (Golden et al., 2024; Griffin et al., 2024). Intersectionality is a way of thinking that underpins the understanding of discipline disparities and how school climate may affect students’ disciplinary outcomes differently. Black youth are socially positioned by a wide range of other characteristics (culture, gender, disability, and ability) that intersect and influence their experiences with exclusionary discipline (Little, & Welsh, 2022; Welsh, 2023). As researchers and policymakers search for ways to disrupt discipline disparities, there remains a need to examine how the relationship between school climate and students’ disciplinary outcomes may vary across student and school characteristics.

This study focuses on the New York City (NYC) context to examine the differential associations between school climate and students’ likelihood of receiving ODRs or suspensions. We focus on a range of student characteristics beyond student race/ethnicity to examine whether school climate improvements benefit certain students more than others. The study is guided by the following research questions:

• RQ1. To what extent is a positive school climate differentially associated with the disciplinary outcomes of students who are Black, Latinx, male, from low-income family backgrounds, with limited English proficiency (LEP), and receiving special education services?

• RQ2. To what extent do school characteristics moderate any differential associations between a positive school climate and the disciplinary outcomes of students who are Black, Latinx, male, from low-income family backgrounds, with LEP, and receiving special education services?

The study builds knowledge and understanding of school climate and school discipline practice and policy. The study empirically responds to Bottiani et al.’s (2017) call for “consideration of appropriate ways to measure school climate to account for the potential influence of differential school discipline in students’ experience of school climate” (p. 11; Bottiani et al., 2017). The resulting insights on whether, for whom, and how a positive school climate may result in a reduction in the likelihood of receiving exclusionary discipline will aid future evaluation, implementation, and exploration studies to better link the dimensions of school climate to students’ disciplinary outcomes. The study provides actionable insights for educational policymakers, researchers, and practitioners on the relationship between school climate and school discipline. Policymakers are encouraged to consider the context of schools such as the diversity of teachers and school leaders as they make investments in improving school climate. The paper also offers researchers areas for developing a richer understanding of how school climate influences racial inequality in school discipline.

School Climate and Students’ Disciplinary Outcomes

Defining and Measuring School Climate

School climate is a complex construct that is widely viewed as an important intervention for school improvement (Berg et al., 2017; Mitchell et al., 2010; Wang & Degol, 2016). Prior studies have established that school climate is associated with a range of academic, socioemotional, and school improvement outcomes (Berkowitz et al., 2017; Cohen et al., 2009; Thapa et al., 2013). For instance, Berkowitz et al. (2017) conducted a comprehensive review of studies on the relationship between school climate and academic achievement and found that more positive school climate is related to better academic achievement for all students and especially for low-socioeconomic status (SES) students. However, school climate can be defined and measured in different ways (see Supplemental Table A1 for a summary of definitions and key domains identified in well-cited prior reviews) (Berg et al., 2017; Berkowitz et al., 2017; Voight et al., 2015). Studies examining school climate have employed varying frameworks to organize the dimensions of school climate (Cornell & Huang, 2016; Huang & Anyon, 2020).

School climate has multiple dimensions, is measured according to the personal experiences of students, parents, and teachers, and reflects the norms, goals, values, relationships, instruction, and organizational structures within schools (Cohen et al., 2009; Thapa et al., 2013). The U.S. Department of Education National Center on Safe and Supportive Learning Environments (2024) defines school climate as “a broad, multifaceted concept that involves many aspects of the student’s educational experience. A positive school climate is the product of a school’s attention to fostering safety; promoting a supportive academic, disciplinary, and physical environment; and encouraging and maintaining respectful, trusting, and caring relationships throughout the school community no matter the setting—from Pre-K/Elementary School to higher education.” and measures 13 subtopics in three main domains (engagement, safety, and environment) in surveys. Cohen et al. (2009) defined school climate as “the quality and character of school life. . . based on patterns of people’s experiences of school life and reflects norms, goals, values, interpersonal relationships, teaching and learning practices, and organizational structures” (Cohen et al., 2009, p. 182). There are four essential dimensions of school climate according to Cohen et al. (2009): safety (physical and social-emotional), teaching and learning (quality of instruction, professional development, leadership, etc.), relationships (diversity, community, and connectedness), and environmental-structural (cleanliness, resources, etc.). In a seminal review of research on school climate, Thapa et al. (2013) identified five essential dimensions of school climate: (a) safety, (b) relationships, (c) teaching and learning, (d) institutional environment, and (e) the school improvement process. Indeed, as highlighted by Berg et al. (2017), “. . .there is growing consensus on the essential components of a positive school climate” (p. 7).

The literature has largely established that school climate influences school discipline, but there is a growing recognition that school discipline may also contribute to school climate (Huang & Anyon, 2020). Students who are suspended tend to have worse perceptions of school climate than their non-suspended peers (Huang & Anyon, 2020), which is consistent with a bi-directional relationship between discipline and climate. For example, the U.S. Department of Education National Center on Safe and Supportive Learning Environments (2023) includes discipline as an aspect of the school environment component of school climate. In Denver, Colorado, discipline permissibility, or whether disciplinary consequences are commensurate with infractions, is an indicator of school climate (Asmar, 2024).

Empirical Evidence on the Relationship Between School Climate and School Discipline

Although conceptually school climate is multidimensional with vast domains and varying definitions, empirically most prior studies linking school climate to students’ disciplinary outcomes have relied on students’ perceptions culled from student surveys as a measure of school climate (Fefer & Gordon, 2020; Gage et al., 2016; Huang & Cornell, 2018; Marraccini et al., 2020; Mitchell & Bradshaw, 2013; Wang & Degol, 2016). Overall, this body of research has largely found that higher rates of out-of-school suspensions and expulsions are associated with more negative perceptions of school climate at the school- and student-level (Gregory et al., 2011; Heilbrun et al., 2018; Mattison & Aber, 2007). Several prior studies have tested hypotheses related to the authoritative school climate theory—which emphasizes disciplinary structure and student support as the salient dimensions of school climate—and found that middle and high schools with structure (i.e., students perceive the rules as fair and strict) and support tend to have lower suspension rates (Cornell & Huang, 2016; Fefer & Gordon, 2020; Gregory et al., 2011; Heilbrun et al., 2018; Huang & Cornell, 2018; Lee et al., 2011). For instance, Gregory et al. (2011) examined the relationship between high school climate and suspensions and found that less structured and supportive schools were associated with higher racial discipline gaps.

Much of the prior work on school climate has been cross-sectional and has not considered changes in school climate over time or the stability of school climate measures (Cohen et al., 2009; Marraccini et al., 2020; Thapa et al., 2013). A limitation of previous studies is that “. . .very few examine school change over time, a key to understanding school improvement processes and efforts” (Thapa et al., 2013, p. 371). Additionally, less attention has been paid to the differential associations of school climate and students’ disciplinary outcomes by student race. Using a cross-sectional sample of middle school students in Virginia, Huang and Cornell (2018) found that even though positive school climate was associated with a decrease in the likelihood of suspensions, the benefits of school climate did not vary differentially by student race/ethnicity. Heilbrun et al. (2018) used student and teacher reported measures of school climate in a cross-sectional sample of middle schools in Virginia and found that only student-reported measures predicted lower suspensions rates, and higher disciplinary structure (i.e., student perceptions of fairer and stricter school discipline policies) was significant for Black but not White students. Gregory et al. (2018) used a cross-sectional student-level dataset for disciplined students (i.e., students who received an ODR) from Denver public schools to investigate the moderation effects of restorative practice interventions on student likelihood of receiving a suspension. They found that restorative interventions significantly reduced the likelihood of student suspension, but the reduction was only slightly larger for Black students compared to White students.

School Racial Climate and School Discipline

Griffin et al. (2024) highlighted the distinctions between school climate and school racial climate. Both examine students’ perceptions of their schooling environment, yet school racial climate studies prioritize “. . .the identification of unique dimensions of the school environment that negatively impact racially-minoritized students’ experiences of the school environment. . .” (p. 4). This entails attention to racism in interpersonal interactions, race relations and stereotypes and racial diversity within schools. School racial climate acknowledges the within-school differences in school climate experiences across student race/ethnicity (Golden et al., 2024; Griffin et al., 2024; Voight et al., 2015). As Griffin et al. (2024) noted “. . .studies have consistently demonstrated that students of different races experienced school climate differently” (p. 3). Black students tend to report more negative perceptions of school climate than White students (Anyon et al., 2016; Konold et al., 2017). Bottiani et al. (2017) found that Black students had more negative perceptions of school equity and school belonging in schools with substantial racial discipline disparities. Voight et al. (2015) found that Black and Latinx students reported lower perceptions of safety and school connectedness than White students, and that the magnitude of these differences varied significantly by school characteristics. Bottiani et al. (2016) assessed school climate by students’ perceptions of support from teachers and found that Black students reported significantly lower levels of perceived support from teachers, even after controlling for SES. Additionally, students who were suspended had worse perceptions of school climate (Huang & Anyon, 2020).

Leveraging cross-sectional data from the Racial Climate Survey administered to students in two Midwestern high schools, Mattison and Aber (2007) found that schools with more positive perceptions of racial climate yielded student reports of higher grades and fewer detentions and suspensions. The authors also found that racial disparities in discipline and grades were also associated with differences in perceptions of racial climate, with White students having more positive perceptions and better disciplinary outcomes than Black students. Golden et al. (2018) used a cross-sectional dataset with 126 Black middle school students in the Midwest found that more positive perceptions of racial climate were generally associated with better self-reported academic outcomes, but this relationship was moderated by peer student perceptions. More recently, Wang et al. (2023) used a 3-year longitudinal dataset with a sample of 1,515 Black and White students in the mid-Atlantic region to investigate the relationship between student perceptions of school racial socialization and suspensions and found that better perceptions of racial socialization were associated with lower risk of suspension. The authors also found that school climate mediates the relationship between racial socialization and discipline, that is, better racial socialization was linked to better school climate which in turn was linked to better discipline outcomes.

Applying a Systems View of School Climate to Differential Associations

We use the Systems View of School Climate (SVSC) (Rudasill et al., 2018) to specify school climate and understand how the components within schools may shape students’ perceptions and disciplinary outcomes. School climate is defined as “the affective and cognitive perceptions regarding social interactions, relationships, values, and beliefs held by students, teachers, administrators, and staff within a school” (Rudasill et al., 2018, p. 46). As such, school climate is distinct from school structures, processes, and context, even though these may shape school climate (Rudasill et al., 2018). The disciplinary process in schools or the sequence from perceived student misbehavior to a disciplinary consequence is an example of a decision-making process in schools that may “be the objective basis of some perceptions that contribute to school climate” (Rudasill et al., 2018, p. 47). Rudasill et al. (2018) incorporates Bronfenbrenner’s (1989) Ecological Systems Theory that posits that students learn in various nested systems—microsystem, mesosystem, exosystem, and macrosystem—that interact with each other. Rudasill et al. (2018) extends this framework to include nanosystems, which are within the microsystem and consist of structures within schools such as classrooms. In this study, we focus on the association between student perceptions of school climate (distinct from school staff or parent perceptions) and students’ disciplinary outcomes (the likelihood of receiving exclusionary discipline). We use SVSC to situate the role of both student and school characteristics (especially the student racial/ethnic composition within schools, school staff experience and racial/ethnic composition, and school-level achievement) in shaping disciplinary outcomes. The likelihood of being referred to the office or being suspended may be influenced by students’ family backgrounds or schools (microsystems) or their teachers or peers in the classroom (nanosystems). In this study, we pay particular attention to the relationship between students’ race, class, and disability status and school climate given that prior studies have highlighted these as central to students’ experiences in schools (Berg et al., 2017; Voight et al., 2015).

Schools, Classrooms, School Climate, and School Discipline

Schools play an important role in contributing to racial inequality in school discipline, and a robust body of studies have highlighted the importance of school-level and classroom-level policies and practices (Girvan et al., 2017; Huang & Cornell, 2018; Osher et al., 2022; Skiba et al., 2014; Welsh & Shafiqua, 2018a; Welsh et al., 2025). Several studies examining the school-level predictors of suspensions have found that the percentage of Black students in a school, school-level poverty, school size and school academic achievement shape the rates of exclusionary discipline within schools (Huang & Cornell, 2018; Welsh & Shafiqua, 2018a; Welsh et al., 2025). Recent studies have also highlighted the importance of school-level factors such as school leader-student racial congruence and the interactions between teachers and school leaders as key considerations in disciplinary outcomes (Welsh, 2023b, 2024). A growing number of studies have underlined the importance of the diversity of teachers for school discipline (Author, 2024; Hughes et al., 2020; Lindsay & Hart, 2017; Shirrell et al., 2023). However, the characteristics of schools that may moderate the benefits of a positive school climate remain underexplored.

Classrooms are nested in schools thus the interactions and relationships in the classroom are an important component of the disciplinary process in schools as well as school climate (Berg et al., 2017; Osher et al., 2022). Classrooms are the sites of many disciplinary infractions at the core of disparities (Austin et al., 2024; Smolkowski et al., 2016) and the majority of ODRs stem from minor behavior issues in the classroom, not violent offenses (Girvan et al., 2017). Prior studies have highlighted the importance of teachers in the disciplinary process (Osher et al., 2022; Rodriguez et al., 2024b) and the centrality of classroom management (Osher et al., 2022; Welsh, 2023b). The characteristics of teachers shape school discipline and school climate in complex ways (Osher et al., 2022; Rodriguez et al., 2024b). Disciplinary moments capture the progression of the teacher-student interactions to disciplinary consequences (Vavrus & Cole, 2002) and Osher et al. (2022) highlighted the importance of racialized classroom interactions. Teachers may implicitly believe that Black students specifically are innately less stable, more likely to misbehave and create trouble, and are more likely to be perceived as a potential risk or threat (Gilliam et al., 2016; Neha & Welsh, 2023; Owens, 2022; Welsh & Shafiqua, 2018a). Prior studies have also highlighted the importance of teacher-student racial congruence in contributing to disciplinary outcomes (Lindsay & Hart, 2017; Shirrell et al., 2023). For example, Lindsay and Hart (2017) found that Black male and female students are less likely to experience exclusionary discipline when they are matched with a Black teacher across grades. The nanosystem within the SVSC framework captures the complexity of the interactions of students and school personnel within classrooms and acknowledges the presence of disciplinary moments and tensions that may shape students’ perceptions (Vavrus & Cole, 2002; Welsh, 2023). As such, this study includes ODRs as a key disciplinary outcome to capture teacher discretion in managing perceived misbehavior in classrooms.

Rudasill et al. (2018) argue that conceptual clarity is necessary for school climate construct validity, suggesting that school climate constructs should focus on any or all of the three components of school climate: (a) perceptions of social interactions and relationships within the school; (b) shared beliefs and values in the school and; (c) the sense of safety within the school. We use two school climate constructs consistent with the above: (a) a peer student behavior construct that measures student perceptions of social interactions and relationships between students, and (b) an overall school environment construct that measures student perceptions of safety and inclusive relationships between school staff and students (more details are provided in the Data and Methods section below). Our school climate constructs also align with the NYCPS conceptualization of school climate (which the NYCPS refers to as “school quality”), which includes six elements of climate (rigorous instruction, collaborative teachers, supportive environment, effective school leadership, strong family-community ties, and trust), with all of the relevant survey questions in our two constructs falling under the NYCPS supportive environment heading. Griffin et al. (2024) called for the application of intersectionality theory to examine the experiences and perceptions of Black girls in schools. This study answers that call with an intentional focus on race, school racial climate survey items and an intersectional lens when examining the relationship between school climate and students’ disciplinary outcomes. In this study, we distinctly explore school racial climate by examining survey items that are linked to perceptions of racial fairness of practices, racial equity, and support. Our analyses are guided by the four themes highlighted by Griffin et al. (2024): (a) interpersonal interactions, (b) fair treatment and racial equity, (c) stereotypes and race relations, and (d) support. Specifically, we include five items related to student perceptions of teacher and curricular inclusion of and respect for different student racial and cultural backgrounds, as shown in Table 2 (see more details in the Data and Methods section below).

Overall, this study builds on the prior research by examining whether the association between positive school climate and better disciplinary outcomes varies by student and school characteristics. It is one of the few studies to investigate school discipline and school climate while controlling for both school and individual student characteristics (Huang & Cornell, 2018). The lack of longitudinal student-level data on school climate and students’ disciplinary outcomes have limited the investigation of the directionality of the effects between school climate and disciplinary outcomes (Huang & Anyon, 2020). Prior studies were also limited to cross-sectional samples and did not account for student-level characteristics or analyze student and school level factors separately in the examination of school climate and disciplinary outcomes (Gregory et al., 2011; Heilbrun et al., 2018; Huang & Cornell, 2018). We employ a rich, longitudinal data from the largest district in the U.S. to examine the overlooked element of the differential associations between school climate and students’ disciplinary outcomes (both ODRs and suspensions) in middle and high schools. In the study design, discussed in further detail in the next section, we leverage consistent items for a longitudinal analysis. Thus, the study focused on survey items that were consistent year-to-year in order to examine the relationship between school climate and school discipline using longitudinal data. We recognize that the longitudinal approach may exclude several survey items that changed across different waves that may capture different dimensions of school climate. As such, we include some analyses with fewer years with a focus on school racial climate and apply an intersectional approach to leverage the breadth of the survey items.

Data and Methods

This study relies on four sources of longitudinal and administrative data provided by New York City Public Schools (NYCPS) for all public middle and high schools spanning the period from 2011–2012 to 2018–2019. The four different sources of data include: (a) infraction-level discipline records, which provide information about the nature of a given reported student infraction along with its disciplinary outcome; (b) student-level administrative data, which include demographic and background information; (c) school-level administrative data, which provide information on a variety of school characteristics; and (d) student responses to the annual New York City School Survey (analyzed at the individual level), which prompts students to answer question items probing their perceptions of the school environment and interactions between peer students and school staff.

Discipline records provide infraction-level information that indicates whether a student received an ODR for a particular infraction. For each infraction, records indicate the severity of the infraction, which ranges from Level 1 (Uncooperative/Noncompliant Behavior) to Level 5 (Seriously Dangerous or Nonviolent Behavior), along with whether that particular infraction led to a suspension for a student. NYCPS distinguishes between two types of suspension—a principal’s suspension and superintendent’s suspension—that differ on the basis of the level of severity of the alleged infraction as well as the duration under which the student is suspended.

The student-level data capture background and demographic information on all students in NYCPS, including their school assignment, grade level, racial and ethnic background (White, Black, Hispanic/Latinx, Asian, or other racial/ethnic background), gender (male or female), socioeconomic status, eligibility for special education services, English proficiency, and temporary housing status.

School administrative records provide a range of data on the nature of school environment and the characteristics of students enrolled in schools including: (a) grade configuration; (b) student enrollment size; (c) the demographic composition of students served (race/ethnicity, socioeconomic status, special education status, English language proficiency and temporary housing status); and (d) measures of school performance (average student attendance, graduation rate, and percent proficient in standardized ELA and mathematics exams). Additionally, human resources data provide demographic and background information for all employed education personnel in a school. Virtually all schools in the available sample are staffed by one principal in a given year, two assistant principals and multiple teachers. For each staff type, we operationalize staff demographics as the percent of staff who are Black or Latinx as well as the percent who identify as male. Average years of experience serves as measures of staff qualifications. While prior research indicates that teacher degree attainment is a poor predictor of student achievement (Hanushek & Rivkin, 2006), the research base has not sufficiently explored its predictive relationship with student discipline outcomes (Rodriguez et al., 2024b); thus, we include the percentage of teachers who received Master’s degree or more than a Master’s degree (above the minimum 30 credits or a doctorate) as an additional measure of staff qualifications. While a sizable proportion of principals and assistant principals are Black (15–17%) and Latinx (27–28%), the demographic makeup of school leaders is notably less diverse than the population of students served in NYC, 31 and 43% of whom are Black and Latinx, respectively.

The NYC School Survey provides the fourth and final source of data for the study, which is administered annually to all students throughout the public school system. The survey is designed to help school leaders better understand the experiences of their students and to diagnose areas to improve the learning environment in their school. We elaborate on the handling of survey information to construct measures of school climate in the following section.

Table 1 provides details on the analytic sample used in the study, including the breakdown by student survey completion status. Almost all of the student and school variables have small but statistically significant average differences between students who completed the survey and those who did not. The student and school characteristics with the largest standardized differences include students receiving special education services, student race (especially students being Black), the proportion of students at a school who receive special education services, percentage of teachers who are male, the racial composition of students within the school, average attendance rate, the graduation rate, and the math and ELA proficiency rates. Based on additional descriptive analyses (not shown in Table 1), Black students and students receiving special education services tend to have worse perceptions of school climate but are less likely to respond to the survey; thus, it is possible that we slightly undersample students with worse perceptions of school climate. We consider how results from the study are robust when reweighting the sample to account for non-response bias in auxiliary specification checks.

Table 1.

Sample Descriptives, All Grade 6 to 12 Students and by Survey Response Status.

Student and school characteristics	All students		Responded to survey	Did not respond to survey	Raw mean difference	Standardized mean difference
	Mean	SD	Mean	Mean
Student-level characteristics
Received ODR	0.13		0.12	0.15	−0.03***	−0.09
Received suspension	0.05		0.04	0.07	−0.03***	−0.12
Black	0.27		0.25	0.33	−0.08***	−0.17
Latinx	0.40		0.39	0.41	−0.02***	−0.04
Asian	0.17		0.19	0.12	0.07***	0.20
White	0.14		0.15	0.11	0.04***	0.11
Other race	0.02		0.01	0.03	−0.02***	−0.10
Female	0.48		0.50	0.43	0.07***	0.14
Low socioeconomic status	0.92		0.94	0.89	0.05***	0.16
LEP	0.15		0.13	0.18	−0.05***	−0.13
Receiving SPED services	0.23		0.18	0.35	−0.17***	−0.39
In temporary housing	0.08		0.07	0.10	−0.03***	−0.10
School-level characteristics
Avg. years teacher experience	10.36	2.81	10.30	10.53	−0.23***	−0.08
Pct. teachers w/MA	99.00	3.20	99.09	98.99	0.10***	0.04
Pct. teachers w/higher than MA	0.51	2.77	0.46	0.57	−0.01***	−0.05
Pct. teachers male	37.55	12.11	36.87	40.25	−3.38***	−0.30
Pct. teachers Black/Latinx	35.85	20.63	29.30	33.01	−3.70***	−0.19
Principal years experience	18.97	5.54	19.07	18.69	0.38***	0.07
Principal w/MA	0.01		1.28	1.83	−0.55***	−0.05
Principal w/higher than MA	0.98		98.18	97.60	0.58***	0.04
Principal male	0.51		50.22	52.62	−2.40***	−0.05
Principal Black/Latinx	0.33		32.48	34.16	−1.67***	−0.04
Avg. years assist. principal experience	16.22	4.33	16.22	16.24	−0.02***	−0.01
Pct. assist. principals w/MA	20.13	32.78	18.05	17.86	0.19***	0.01
Pct. assist. principals w/higher than MA	79.48	32.94	81.53	81.50	0.03	0.00
Pct. assist. principals male	38.74	37.44	39.78	41.17	−1.39***	−0.04
Pct. assist. principals Black/Latinx	44.38	41.60	33.94	39.29	−5.35***	−0.15
Student-teacher ratio	18.03	8.45	17.77	18.79	1.02***	−0.12
Student enrollment	1,272.52	1,205.75	1,289.32	1,233.65	55.67***	0.05
Pct. students Black	27.46	44.63	25.63	31.53	−5.90***	−0.25
Pct. students Latinx	39.80	48.95	39.46	40.33	−0.87***	−0.04
Pct. students Asian	16.77	37.36	18.09	13.61	4.48***	0.26
Pct. students other race	1.81	1.33	1.43	2.67	−1.25***	−0.22
Pct. students low socioeconomic status	92.38	26.53	80.00	79.75	0.25***	0.01
Pct. students in temporary housing	7.68	26.62	6.43	7.13	−0.69***	−0.12
Pct. students receiving SPED services	23.05	42.12	16.69	27.48	−10.80***	−0.53
Pct. students LEP	14.96	35.67	12.47	14.05	−1.58***	−0.11
Avg. attendance rate	0.87	0.11	0.90	0.83	0.07***	0.57
Graduation rate	0.77	0.18	0.79	0.71	0.07***	0.40
Pct. students proficient in math	10.65	18.46	13.84	5.20	8.64***	0.50
Pct. students proficient in ELA	11.56	17.98	14.62	5.62	9.00***	0.51
N (student-years)	3,988,302		2,911,460	1,076,842
Unique student sample size	1,167,306		852,133	315,173
N (school-years)	7,685
Unique school sample size	1,875

Note. Standard deviations shown for continuous measures only. SD = standard deviation; ODR = office discipline referral; LEP = limited English proficiency; SPED = special education; MA = Master’s degree; ELA = English language arts.*p < .05. **p < .01. ***p < .001.

Key Constructs

Discipline Outcomes

Raw discipline records were made available at the infraction level, however, the study’s unit of analysis is at the student-year level. Therefore, infraction-level discipline information is aggregated to the student-year level to operationalize two binary indicators for the outcomes of interest: (1) a binary indicator for whether a student received at least one ODR (Level 1 through Level 5) and (2) a binary indicator for whether a student received a suspension (principal or superintendent) in a given year. We analyze the principal and superintendent suspension jointly given that the overwhelming majority of suspensions received by students comprise principal suspensions (73–81% within a given school year).

School Climate

For the main purpose of the study, we focused on 11 survey items that were consistently available across the entire 8-year data panel. To reduce the dimensionality of the survey data, we conduct an exploratory factor analysis (EFA) for student survey items. Based on visual examination of the scree plot and the Kaiser–Guttman stopping criteria, we retained two factors based on student survey items representing (1) students’ perception of peer student behavior within school; and (2) students’ perception of the overall school environment. Both factors were standardized to have a mean of zero and standard deviation of one within the sample. Higher values of the peer student behavior and school environment factors indicate a better school climate, with values below zero indicating a worse than average school climate and values above zero indicating a better than average school climate.

The peer student behavior measure is based on student responses to four questions about the frequency of other student behavior, using a 4-point Likert frequency scale (“Most of the time,” “Some of the time,” “Rarely,” and “None of the time”): harassment, bullying and intimidation; physical fights; use of alcohol, illegal drugs, and prescription drug abuse; and gang activity. This measure of school climate encompasses the extent to which students’ assess their peers’ behavior as positive.

The school environment measure is based on student responses to seven questions about the school environment, also using a four-point Likert agreement scale (“Strongly disagree,” “Disagree,” “Agree,” and “Strongly agree”): inclusion of students with disabilities; variety of programs and classes to keep students interested; students treat each other with respect; students feel safe in class; students feel safe in halls, bathrooms, etc.; students feel safe around the school; and school cleanliness. This construct captures safety, cultural awareness and inclusivity dimensions of students’ perception of school climate. Three of the survey questions address perceptions of safety in and around schools. The remaining items can be categorized as how students perceive the cultural awareness and inclusivity of the school environment. The school environment measure bears some similarity to the overall safety construct used in prior studies testing hypotheses related to the authoritative school climate theory (Fefer & Gordon, 2020; Gregory et al., 2011; Heilbrun et al., 2018; Huang & Anyon, 2020; Huang & Cornell, 2018; Lee et al., 2011), which provides support for the validity of this construct as a measure of school climate.

To assess the reliability of our two constructs, we calculate the scale reliability coefficients. Both factors have a high scale reliability coefficient (0.78 for the student behavior construct and 0.82 for the school environment construct) indicating internal consistency of the underlying items.

Table 2 illustrates the survey items in the school climate constructs, along with their corresponding EFA diagnostics. As shown, all items suitably load onto their respective factor, with the minimum loading being 0.55.

Table 2.

Exploratory Factor Analysis (EFA) of NYC School Survey Measures.

Construct	Student survey item	Loading	Eigenvalue	Pct. variance explained	Scale reliability coefficient
(1) Risky student behavior within school	Students harass, bully, or intimidate other students	0.78	2.36	59.02	0.78
	Students get into physical fights	0.69
	Students drink alcohol, use illegal drugs, or abuse prescription drugs	0.78
	There is gang activity	0.82
(2) Overall school environment	Students with disabilities are included in all activities	0.60	3.68	53.14	0.82
	Variety of programs, classes, and activities to keep me interested	0.70
	Most students treat each other with respect	0.55
	I feel safe in my classes	0.84
	I feel safe in the hallways, bathrooms, locker rooms, and cafeteria	0.85
	I feel safe outside around this school	0.81
	This school is kept clean	0.66
(3) Racial climate	My teachers use examples of students’ different cultures/backgrounds/families in their lessons to make learning more meaningful for me.	0.69	3.04	60.85	0.81
	I see people of many races, ethnicities, cultures, and backgrounds represented in the curriculum.	0.74
	In general, my teachers treat students from different cultures or backgrounds equally.	0.81
	I feel that my teachers respect my culture/background.	0.83
	In general, my teachers present positive images of people from a variety of races, ethnicities, cultures, and backgrounds.	0.81
(4) Bullying	At this school students harass, bully, or intimidate each other because of their race, religion, ethnicity, national origin, or citizenship/immigration status.	0.90	2.40	79.98	0.87
	At this school students harass, bully, or intimidate each other because of their gender, gender identity, gender expression, or sexual orientation.	0.90
	At this school students harass, bully, or intimidate each other because of other differences, like disability or weight.	0.89

Note. The “Risky Student behavior within school” and “Overall school environment” constructs are based on all years of survey item responses data in the sample (2011–2012 to 2018–2019), while the “Racial climate” and “Bullying” constructs are based only on a subset of years (2016–2017 to 2018–2019) for which all of the survey item responses data for these constructs are available.

In specification checks, we leverage an expanded set of survey items that were available in only a subset of years to further distinguish key elements of school climate commonly identified by the various conceptualizations of school climate. We conducted an EFA using eight survey items that were consistently available between the years 2017 to 2019 (with five of the items in the expanded set not all being available for the earlier years 2012 to 2016). The results of the EFA yielded two additional factors that included items representing students’ perceptions of racial climate and students’ perceptions of bullying. The racial climate construct is based largely on items about teacher practices relating to student racial backgrounds, while the bullying construct is based on various items about student bullying, intimidation, and harassment. Similar to the other school climate constructs, details of these constructs with their EFA diagnostics are also shown in Table 2. Results generated by using these alternative school climate measures with only 3 years of data are described in the specification checks section.

Analytic Strategy

There are several ways to measure discipline disparities, or the differences in disciplinary outcomes across groups of students. Prior studies have highlighted the strengths and weaknesses of measures such as the relative risk ratio (RRR), absolute risk difference (ARD), and raw differential representation (RDR), and urged researchers to use multiple measures (Curran, 2020; Girvan et al., 2019; Larson et al., 2019; Osher et al., 2015; Rodriguez & Welsh, 2022). Although the RRR is used by the federal government and several state governments, as Osher et al. (2015) highlighted, it has limitations when based on small numbers and may not fully capture progress over time, “For example, (a) in schools with very high disparities, small changes in the comparison group can dramatically change the risk ratio and (b) risk ratio can increase even when the risk gap improves” (p. 35). As such, we also consider the ARD (or the risk gap) as this is a better metric to track disparities over time; however, we are mindful that “the risk gap is not usually useful in examining extreme but low-incident events” (p. 35, Osher et al., 2015). As highlighted by Curran (2020), “To the extent that stakeholders have evidence that overall rates of discipline are decreasing, RD or RDR may be a more desirable measure of disproportionality as these measures are not as susceptible to showing increasing disparities as a function of decreases in underlying discipline rates” (p. 384). Girvan et al. (2019) noted that the RDR provides “. . .complimentary and perhaps critical information about disproportionality in the schools that is not captured by the risk ratio” (p. 44). It is important to highlight that these disproportionality metrics are typically calculated at the school-level whereas the analyses in the current study are at the student-level. Notwithstanding, we pay attention to the changes in predicted probabilities in absolute and relative terms in the interpretation of results.

The analyses rely on logistic regression-based methods to estimate the predictive relationship between student perceptions of school climate, student and school characteristics, and student-level disciplinary outcomes. Using student-level data comprising the analytic sample of middle and high school students, we consider a model of the following form interacting student subgroups of interest with school climate in order to investigate differential relationships:

$$\begin{array}{l}Pr{\left(D=1\right)}_{ist}={\beta }_0+SchCli{m}_{ist}{\Phi }_1+StuCha{r}_{ist}{\Phi }_2+SchCli{m}_{ist}\times \\ StuCha{r}_{ist}^k{\Phi }_3+SchCha{r}_{st}\Theta +{\zeta }_s+{\gamma }_t+{\epsilon }_{ist},\end{array}$$ (1)

where the probability of student i in school s is disciplined during year t is modeled as a function of students’ perception of school climate, observable student characteristics, and time-variant school characteristics. We estimate equation 1 separately for each of the two forms of exclusionary discipline of interest: (a) whether a student received an ODR, and (b) whether a student received a suspension (i.e., principal’s or superintendent’s suspension) in a given year. The specified model controls for observable student characteristics (student race/ethnicity, gender, socioeconomic status, special education status, LEP, and temporary housing status). We measure prior-year disciplinary history using two binary variables at the student-level (student received ODR and student received suspension in the previous year). Time-varying school characteristics include percent of students who are Black/Latinx; percent of students who are Asian; percent of students who are other race; percent students proficient in ELA/math; school average attendance rate; percent students from a low socioeconomic background; percent students in temporary housing; percent student receiving special education services; percent students with LEP; teacher, principal, and assistant principal race/ethnicity, gender, education and experience level; ratio of students to teachers; and student enrollment.

In addition, the specified model adjusts for school fixed effects ( ${\zeta }_s$ ) to control for any time-invariant characteristics affecting the likelihood of students experiencing exclusionary discipline (i.e., school grade configuration, school location) and year fixed effects ( ${\gamma }_t$ ) to adjust for systemic trends in exclusionary discipline affecting the entire NYC public school system. One concern with regard to the adjustment of school fixed effects is that school climate may be a fairly stable characteristic within schools over time. Much of the prior work on school climate has been cross-sectional and has not considered changes in school climate over time or the stability of school climate measures (Cohen et al., 2009; Marraccini et al., 2020; Thapa et al., 2013). An analysis of variance in a school-year data set revealed that over 78% of the total variation in students’ assessment of “Peer Student Behavior” and 70% of the total variation in the “Overall School Environment” measure is within schools over time.

In equation 1, $SchCli{m}_{ist}$ represents a vector of measures for student perception of school climate. The vector coefficient ${\Phi }_1$ , therefore, would predict the isolated difference in the likelihood a student experiences a form of exclusionary discipline that is associated with variation in students’ perception of school climate. A negative and statistically significant value for an element within the ${\Phi }_1$ vector parameter would be indicative that a more positive school climate is associated with students being less likely to experience a form of exclusionary discipline.

The main question of interest, however, is to understand the extent to which school climate differentially predicts the likelihood of experiencing exclusionary discipline for specific subgroups of students (RQ1). We specifically explore the extent to which students’ perception of school climate differentially predict differences in student exposure to discipline for students based on race/ethnicity, gender, socioeconomic status, temporary housing status, LEP status, and special education status. The subgroup of interest is specified by a given student characteristic k (i.e., race/ethnicity, gender, socioeconomic status, temporary housing status, LEP status, special education status), which is represented by element k of the $StuCha{r}_{ist}$ vector term. Here, the vector parameter of interest is ${\Phi }_3$ , which captures the extent to which school climate differentially predicts the probability of experiencing exclusionary discipline for students within the subgroups of interest specified by $StuCha{r}_{ist}^k$ .

In order to address the second research question, we modify equation 1 to include three-way interactions between the school climate constructs ( $SchCli{m}_{ist}$ ), student characteristic of interest ( $StuCha{r}_{ist}^k$ ), and school characteristics of interest, including the student racial/ethnic composition within schools, school staff experience and racial/ethnic composition, and school-level achievement. To facilitate interpretation of results, we present our findings using visual display of predicted probabilities for significant interactive relationships (with regression output shown in Supplemental Appendix tables); however, the full set of estimated regression coefficients are available upon request from the authors. For the purposes of the predicted probabilities displayed, we show school characteristics of interest as binary variables (above or below median percent students Black or Latinx; school is low-achieving if school is in lowest tercile for average of ELA and Math proficiency rates; principal is Black or Latinx; above/below percent assistant principals Black or Latinx; and above/below percent teachers Black or Latinx).

We also run several specification checks for each of the above models. First, we consider if there is a differential relationship between school climate and discipline outcomes based on student intersectional identities, by re-running our models with double interactions between climate and student race and gender, and then triple interactions between climate and student race, gender, and special education status, with particular interested in student groups most at-risk of receiving exclusionary discipline: Black male students and Black male students receiving special education services. Second, despite the robust student response rates to the NYC School Survey, we re-estimate our analyses using a propensity score adjustment weighting approach to account for potential non-response bias (Valliant & Dever, 2018). This approach involves modeling the propensity that individual students responded to the survey based on their observable characteristics and then weighting each observation by the inverse of its estimated propensity score so that responding students with low propensity scores (i.e., those who are predicted to be less likely to respond to the survey based on their observed characteristics) are upweighted in the reweighted sample to adjust for non-responding students. Third, we predict student likelihood of ODR and suspension in 1 year from the two measures of school climate in the year prior (a lagged model).

Results

Who Benefits From a Positive School Climate?

First, we present results on which students’ disciplinary outcomes benefit from positive school climates. Both the peer student behavior and school environment climate constructs have significant negative associations with both the likelihood of receiving an ODR and suspension. In other words, more positive school climates (as measured by higher student perceptions) are associated with better disciplinary outcomes at the student-level, after controlling for student and school characteristics, prior student disciplinary history, and school and year fixed effects. The significant interactions are shown in graphical form as predicted probabilities in Figures 1 to 5 (Supplemental Tables A2–A6 in the Appendix show the results for these logistic regression models as odds ratios, with values below 1 indicating a lower likelihood of receiving ODR/suspension and values above 1 indicating a higher likelihood).

Figure 1.

School climate and probability of receiving suspension, by student race, socioeconomic status, and gender: (A) probability of suspension and peer student behavior, by student race, (B) probability of suspension and peer student behavior, by student socioeconomic status, and (C) probability of suspension and peer student behavior, by student gender.

Figure 2.

School climate and probability of receiving ODR or suspension, by school racial composition and student race: (A) probability of suspension and peer student behavior, by student race and school racial composition and (B) probability of ODR and school environment, by student race and school racial composition.

Figure 3.

School climate and probability of receiving ODR or suspension, by school achievement and student race: (A) probability of suspension and peer student behavior, by student race and school achievement and (B) probability of ODR and school environment, by student race and school achievement.

Figure 4.

School climate and probability of receiving ODR, by school principal, assistant principal, teacher, and student race: (A) probability of ODR and school environment, by student race and teacher racial diversity, (B) probability of ODR and school environment, by student race and principal racial diversity, and (C) probability of ODR and school environment, by student race and assistant principal racial diversity.

Figure 5.

School climate and probability of receiving ODR or suspension, by school achievement and student gender, socioeconomic status, and SPED-status: (A) probability of suspension and peer student behavior, by student gender and school achievement, (B) probability of ODR and school environment, by student socioeconomic status and school achievement, and (C) probability of ODR and peer student behavior, by student SPED status and school achievement.

There are significant differential relationships for five of the six student characteristics of interest—student race, gender, socioeconomic status, SPED status, and temporary housing status—and these significant differential relationships exist regardless of the school climate construct (peer student behavior or school environment) or the disciplinary outcome (ODR or suspension) considered. There is no significant differential association for students based on LEP status. Figure 1 illustrates examples of the predicted probabilities from the two-way interactions of student characteristics and school climate for suspension and peer student behavior, by student race (Plot A), socioeconomic status (Plot B), and gender (Plot C). Patterns are similar for ODR, though not shown in Figure 1.

As illustrated in Figure 1—based on estimates presented in supplementary Table A2—students across all racial/ethnic backgrounds (Asian, Black, Latinx, and White) are less likely to receive ODRs and suspensions when enrolled in schools with a more positive climate for both constructs. However, the reduction in the likelihood of ODRs and suspensions is statistically significantly smaller for Black and Latinx students relative to White students. Specifically, as the peer student behavior climate measure increases from −3 (representing an extremely negative school climate) to 3 (an extremely positive school climate), the predicted probability of suspension decreases in the absolute sense by 4% for Black students, 4% for Latinx students, and 3% for White students; but in the relative sense, this reduction is larger for White students when compared to their starting point (i.e., predicted probability of suspension in an extremely negative school climate) which was much lower than for Black and Latinx students. This differential reduction by race/ethnicity in response to improvements in school climate suggests changes in disciplinary disparities across groups.

To further assess these changes, we consider two metrics: the absolute difference and the relative difference in the likelihood of receiving an ODR or suspension across racial/ethnic groups (Curran, 2020; Girvan et al., 2019; Osher et al., 2015; Rodriguez & Welsh, 2022). Similar to Larson et al. (2019), the results using the relative differences diverge from those using the absolute difference. The absolute difference in predicted probabilities slightly declines as the peer student behavior climate measure improves. For instance, the absolute gap in the predicted probability of suspension between Black and White students is 4% when the peer student behavior is −3 and 3% when it is 3. One may infer that more positive school climates are associated with a reduction in racial disparities in exclusionary discipline using absolute difference as the measure of disproportionality. In contrast, the relative gap increases with a more positive school climate. The predicted probability of suspension is 1.6 times higher for Black students compared to White students when the climate is −3, but this ratio increases to 2 times higher when the climate is 3.

The differences in results between relative differences and absolute differences partly occurs because in extremely negative school climates, the starting probability of suspension for Black students (almost 8%) is much higher than for White students (less than 5%). Consequently, the relative reduction in the probability of suspension is smaller for Black students than for White students. Therefore, while the absolute difference in the predicted probability of receiving an ODR or suspension between Black and White students decreases slightly, the relative difference increases as the school climate becomes more positive.

To provide some context for the above results about school climate and discipline disparities, we consider the RDR in schools with more negative or positive school climates. In the average school with approximately 1,273 students including 344 Black students, the RDR (or the number Black students who experienced suspension but would not have if they were subject to discipline at the same rate as White students) is 14 when the peer student behavior is −3 and 11 when the peer student behavior is 3. In other words, improved school climate is only associated with a very small reduction in the number of Black student suspensions in a typical school, in terms of the raw number of students. Applied to the entire population of students in our sample across all NYC public middle and high schools, the RDR is 19,942 when the peer student behavior is −3 and 14,956 when the peer student behavior is 3 (a meaningful difference).

The differential associations for Asian students and other race students are less clear. Asian students have a significantly larger reduction in likelihood of ODR and suspension in schools with better peer student behavior compared to White students; however, for the school environment construct, the opposite relationship (a smaller reduction in likelihood compared to White students) is found for student ODR and no significant differential relationship is found for student suspension. The disciplinary outcomes of other race students appear to have no significant differential association with school climate, except for having a smaller reduction in risk of suspension associated with a more positive school environment than White students.

There is also evidence of a differential relationship between school climate and discipline outcomes based on gender. Female students appear to have a significantly larger reduced risk of receiving exclusionary discipline associated with more positive school climates than male students, and this is the case for both school climate measures and both ODRs and suspensions. The results also indicate that students who are economically disadvantaged (low socioeconomic status or in temporary housing) or receiving special education services have a significantly smaller reduced risk of ODR and suspension associated with more positive school climates than their more advantaged peers. Overall, the results highlight variation in how different students’ disciplinary outcomes are associated with measures of school climate using students’ perceptions.

Do School Attributes Moderate the Association Between School Climate and Disciplinary Outcomes?

Next, we turn our attention to whether school characteristics moderate the differential relationships between disciplinary outcomes and a positive school climate. Specifically, we focus on (a) the racial diversity and experience of school leadership (both principals and assistant principals) and teachers; (b) the racial composition of schools; and (c) the academic achievement of schools. We find evidence that school attributes moderate the association between school climate and disciplinary outcomes and that this varies mostly by student race.

Figure 2 illustrates the predicted probabilities from triple interactions of student characteristics, school climate constructs, and school racial composition. School racial composition significantly moderates the relationship between school climate and the disciplinary outcomes of Black and Latinx students. As shown in Plot (A), more positive perceived peer behavior is associated with a lower probability a student receives a suspension, however, the strength of relationship is moderated by the race of the individual student as well as the racial composition of students in that student’s school. On average, Black students tend to have the highest probability of receiving a suspension; however, the predicted relationship between perceived peer behavior and the probability of receiving a suspension is significantly weaker for Black students compared to all other students, particularly for Black students who are in schools that are serving an above-median percentage of students who are Black or Latinx. Similarly, the predicted probability of receiving an ODR tends to decline for all student groups as perceived school environment becomes more positive. However, as shown in Plot (B), the predicted marginal change in the probability of receiving an ODR is weaker for Asian, Black, and Latinx students compared to their White peers in schools serving a below-median percentage of students who are Black or Latinx, but stronger for those groups in schools serving an above-median percentage of students who are Black or Latinx. The same heterogeneous pattern is observed between perceived school environment and probability of suspension (not shown).

Figure 3 illustrates the predicted probabilities from triple interactions of student characteristics, school climate constructs, and school achievement. As Plot (A) shows, more positive perceived peer student behavior is associated with a lower probability of receiving a suspension in non-low-achieving schools; however, the predicted decline is smaller in magnitude for Black students compared to all other student groups. In low-achieving schools, the predicted decline in the probability of receiving a suspension associated with more positive peer student behavior is significantly attenuated for non-Black students. This same heterogenous relationship is observed for perceived school environment and probability that students receive an ODR (Plot B) and suspension (not shown).

Figure 4 shows the predicted probabilities from triple interactions of student characteristics, school climate constructs, and the diversity of school personnel such as teachers and school administrators. More positive school environment is associated with a lower probability of receiving an ODR. In schools with a below-median share of teachers who are Black or Latinx, the predicted decline in the probability of receiving an ODR associated with a more positive school environment is significantly greater for White students compared to Asian, Black, and Latinx students (Plot A). However, in schools with an above-median share of teachers who are Black or Latinx, the predicted decline in the probability of receiving an ODR associated with a more positive school environment is more evenly observed across all students. The same heterogeneous pattern is observed between perceived school environment and probability of ODR when comparing school contexts based on principal race/ethnicity (Plot B) and assistant principal race/ethnicity (Plot C).

Figure 5 illustrates how school achievement moderates the differential association between school climate and students’ disciplinary outcomes by student gender, socioeconomic status, and SPED-status. As shown in Plot A, in schools that are not low-achieving, as perceived peer student behavior becomes more positive, the risk of receiving or suspension is predicted to decline for female students to a greater degree than for male students; however, this differential association is attenuated in low-achieving schools. This same differential pattern is observably significant for the risk of receiving an ODR as well (not shown), although these interactions are not significant when the school environment climate measure is used in lieu of perceived student behavior.

In contrast, school characteristics do not appear to moderate the association between perceived peer student behavior and disciplinary outcomes by student socioeconomic status; however, the results vary with the use of school environment as the measure of school climate (Plot B). As perceived school environment becomes more positive in non-low-achieving schools, students from a low socioeconomic background have a smaller reduced risk of ODR relative to other students; however, this differential association is attenuated in low-achieving schools. Similar differential results are found for students from a low socioeconomic background in schools with above median percent teachers Black or Latinx (not shown in figures).

We find very little evidence that school characteristics moderate the differential association between school climate and ODR and suspension by student’s special education status. The exception is that as perceived peer student behavior becomes more positive in non-low achieving schools, the relative difference in the probability of receiving an ODR declines between students who are receiving special education services and those who are not (although the absolute difference in the probability of receiving an ODR increases). However, the change in the relative difference across both student groups is attenuated in low-achieving schools (Plot C).

We also find evidence that school racial composition and teacher diversity moderate the relationship between school climate and discipline outcomes by student gender (though not shown in the figures). Male students have a significantly larger reduction in risk of suspension and ODR associated with a more positive school climate (using both school climate measures) in schools with more student and teacher racial diversity than female students. Principal race and experience, assistant principal race and experience, and teacher experience do not appear to moderate the association between school climate and discipline outcomes by student gender, socioeconomic status, or SPED-status.

Robustness and Sensitivity Checks

We perform several model specification and sensitivity checks to assess the robustness of our results. First, we aggregate student perceptions of school climate across both constructs to determine schoolwide perceptions. We also aggregate perceptions for comparison by various student groups—specifically, by race, gender, socioeconomic status, temporary housing status, SPED status, and LEP status. This allows us to examine how aggregated perceptions of climate within these groups relate to students’ disciplinary outcomes (e.g., how the average Black student perception of school climate relates to a Black student’s likelihood of receiving ODR or suspension). We find consistent evidence of larger significant differential associations between disaggregated school climate and disciplinary outcomes for each student group, compared to the schoolwide perceptions of school climate. The results indicate that students who are Black, male, low-income, or receiving special education services have larger reductions in likelihood of receiving ODRs and suspension associated with more positive school climates as measured by the perceptions for just their student group. All of the significant triple interactions found above are still significant with the models including disaggregated measures of school climate by student group. This provides some evidence that school climate and perceptions of school climate varies across student groups—which may explain part of the differential associations found—but schoolwide climate constructs still help explain disciplinary outcomes. However, the aggregated perceptions of climate (schoolwide and by student group) have a weaker relationship with disciplinary outcomes than the individual student perceptions of climate used in the main analyses.

In order to investigate how school climate may have a differential relationship with disciplinary outcomes based on student intersectional identities, we run separate analyses with double interactions between climate and student race and gender, and triple interactions between climate and student race, gender, and special education status. Overall, we find that the students with the highest ODR rates (e.g., Black male students with disabilities) have the weakest relationship between school climate and disciplinary outcomes, even though school climate (using both student factors, for both ODR and suspension) is related to lower likelihood of exclusionary discipline for all student groups. There is a significantly stronger relationship between school climate (as measured by the student behavior construct) and the likelihood of receiving ODR for Black female students (compared to Black male students), Black students not receiving special education (compared to Black students who do), and Black students who are not either male or receiving special education (compared to Black males receiving special education). Supplemental Figure A1 in the Appendix shows that the relationship between school climate and school discipline is stronger for Black female students than it is for Black male students, and that Black–White racial disparities become worse for males and better for females in better school climates, in both an absolute and relative sense. Panel A of Supplemental Appendix Figure A2 shows that while Black–White racial disparities seem to decline as peer student behavior improves for non-SPED students, racial disparities actually increase for SPED students, for both male and female SPED students. In contrast, Panel B shows that as school environment improves, Black–White racial disparities slightly improve for female non-SPED students but slightly worsen for male non-SPED students (with likelihood of ODR declining for all students), while for SPED students the likelihood of ODR does not appear to decline for any student group and racial disparities appear to increase for both males and females. There are similar patterns for suspension, but unlike ODR most of these interactions are not statistically significant. We do not find any evidence of significant double or triple interactions for the moderation analyses.

Next, we re-run our models, this time predicting ODR and suspension in a given year using the two measures of school climate in the year prior (a lagged model). While most of our results remain largely unchanged, we find slightly larger and more significant differential associations between school climate and student race, suggesting that changes in school climate may be more strongly associated with changes in disciplinary outcomes in subsequent years than in the same year. What we present in the main findings appears to be a lower-bound estimate.

We focus on school racial climate and bullying, using only 3 years of data instead of the full panel of 8 years due to limited availability of some of the constructs. These results are shown in Supplemental Appendix Tables A7 to A11. We find similar overall results: student perceptions of both school racial climate and bullying are related to lower likelihood of ODR and suspension, albeit with less strong relationships compared to our peer student behavior and school environment constructs. We find similar differential relationships to before (i.e., a stronger relationship between discipline outcomes and school climate for female students than for male students, for non-SPED students than for SPED students, and for students not from a low socioeconomic background than for students from a low socioeconomic background), with the exception that for school racial climate there is a less strong differential relationship between climate and discipline outcomes for Black and White students (i.e., White students still seem to benefit most from better school racial climates, but to a lesser degree compared to peer student behavior or school environment). For our moderation models, we find that some school characteristics (school racial composition and teacher diversity, but not school achievement) still appear to moderate the relationship between school racial climate and discipline outcomes. However, we do not find that any of these triple interactions for our moderation analyses retain significance for the student bullying construct.

Finally, we re-estimate our analyses using a propensity score adjustment weighting approach to account for potential non-response bias. Almost all of our significant results remain significant. However, none of the significant results for Asian or other race students (for either research question) retain significance after making this adjustment. There is also a slightly larger difference between Black and White students in terms of the strength of the relationship between school climate and discipline (White students seem to have an even stronger association between climate and discipline, relative to Black students, compared to the main models).

Discussion

Who benefits from school discipline reforms and whether Black students (who are punished the most (Welsh, 2022; Darling-Hammond & Ho, 2024; Neha & Welsh, 2023; Rodriguez & Welsh, 2022; Welsh & Shafiqua, 2018a; Welsh & Rodriguez et al., 2024c)) differentially benefit from school discipline reforms is an important consideration that partly determines whether interventions reduce the prevalence of exclusionary discipline but not disparities in students’ disciplinary outcomes. As highlighted by school discipline expert Gregory et al. (2021), for school discipline reforms such as school climate and social-emotional learning (SEL), “. . .it remains unclear if SEL programs and interventions are more effective with certain populations than others (e.g., age, ethnicity, presence of risk factors)” (p. 4). This study offers insights on the differential relationships between school climate and school discipline, and as such, provides a richer understanding of which students’ disciplinary experiences are most strongly associated with a positive school climate and in which schools.

Overall, we find that better school climate (regardless of which climate construct we use) is associated with reduced likelihood of both student ODR and suspension, across all student groups, even after controlling for student and school characteristics and students’ prior disciplinary outcomes. The significance of the results after we control for students’ prior disciplinary history and if we predict disciplinary outcomes using prior year school climate, provide some evidence that changes in school climate are associated with subsequent changes in disciplinary outcomes. This highlights the importance of improving school climate to reduce the use of exclusionary discipline. Additionally, our study shows that the association between school climate and school discipline varies significantly by student and school characteristics. The results illustrate that Black, Latinx, male, low-income, special education, and students in temporary housing experience a smaller reduction in the relative likelihood of receiving ODRs and suspensions when they are in schools with more positive climates, compared to other students.

The differential association between school climate and students’ disciplinary outcomes provides an empirical illustration of the importance of using multiple measures of discipline disproportionalities and how the choice of disparities metrics shapes conclusions. As such, we are cautious in the interpretation of relative differences to capture trends in disparities over time given the aforementioned concerns. Notwithstanding, as scholars have highlighted (Curran, 2020; Girvan et al., 2019; Larson et al., 2019; Osher et al., 2015; Rodriguez & Welsh, 2022), each measure of discipline disproportionality tells another dimension of the overall story. In this case, using student-level analyses, the results using relative differences diverge from those using absolute differences when discipline disparities are considered. The implication is that even though a more positive school climate does reduce disciplinary actions for all students, the relative disparity may widen. This happens because the positive association of an improved school climate is less pronounced for higher-risk students (e.g., Black, Latinx, special education status students) compared to their peers. In other words, even though the absolute gap in disciplinary rates might decrease slightly (i.e., the difference in suspension rates between Black and White students narrows), the relative gap often increases. This pattern is particularly evident when we examine how the likelihood of receiving ODRs and suspensions for Black students changes relative to White students as the school climate improves. We further explore this issue by using another measure of disproportionality, the RDR (indicating the number of Black students who are suspended but would not be had they been disciplined at the same rate as White students), and the results suggest that improved school climate is only associated with small RDR reductions at the school level, but at the city level there is a large RDR reduction.

Our findings build on a limited number of prior studies on the differential association between school climate and disciplinary outcomes. Huang and Cornell (2018) found that a more positive school climate as reported by students is associated with lower likelihood of student suspension, but the benefits of a better school climate in terms of disciplinary outcomes do not vary by race. This contrasts with our finding that Black and Latinx students seem to benefit relatively less from more positive school climates compared to White students. Heilbrun et al. (2018) found that schools with more authoritative disciplinary structures as measured by student surveys tend to have lower suspension rates and lower racial disparities in discipline. Our results are consistent with Heilbrun et al. (2018) insofar as better school climate is associated with lower risk of suspension. However, we find that Black and Latinx students have a smaller benefit associated with a more positive school climate (as measured by the RRR) whereas Heilbrun et al. (2018) found that racial discipline disparities (as measured by the ARD) are lower in schools with more positive climates. Gregory et al. (2018) found that school use of restorative justice interventions (that results in better school climate) was associated with overall reductions in student likelihood of suspension (given receipt of ODR) and slightly larger reductions for Black students. Although our results are consistent with Gregory et al.’s (2018) finding that better school climate is linked to reduction in suspension risk, we find that White students have a larger reduction than Black students. Our differences in results compared to these prior studies can possibly be explained by our use of longitudinal data and controlling for student disciplinary history.

The Importance of School Context

Similar to recent studies (Gregory et al., 2021; Osher et al., 2022; Rodriguez et al., 2024b; Welsh et al., 2024a, 2025), the findings underscore the importance of school context (SVSC situates the demographic composition of personnel and students as context) for students’ disciplinary outcomes. Prior studies have found that the proportion of Black students in a school is predictive of likelihood of receiving exclusionary discipline (Welch & Payne, 2010; Welsh & Shafiqua, 2018a; Welsh et al., 2025). Similar to demographic composition, academic achievement is also a significant school-level variable in school discipline (Skiba et al., 2014; Welsh & Shafiqua, 2018a). We find that school racial composition, school achievement, and the diversity of teachers and school leaders moderate the association between school climate and disciplinary outcomes for Black and Latinx students. School characteristics moderate the association between school climate and school discipline more so by student race than other student characteristics such as socioeconomic or LEP status.

In better climates, Black and Latinx students have a greater reduced likelihood of exclusionary discipline compared to White students if they are in schools with higher proportions of Black and Latinx students, low academic achievement, and more staff racial diversity (although some of these interactions are not significant depending on the school climate or exclusionary discipline measure). Black students in low-achieving schools have a larger associated decrease in likelihood of being suspended in schools with more positive school climate, relative to other students. Similarly, in low-achieving schools, improvements in school climate are associated with reductions in likelihood of exclusionary discipline for male students, students from a low socioeconomic background, and students receiving special education, without increasing disparities. Additionally, Asian, Black, and Latinx students in schools with more diverse teachers and school leaders have a larger associated reduction in risk of ODR in schools with more positive school climate, compared to their peers. In essence, teacher and school leader (principals and assistant principals) diversity moderates the disciplinary benefits of school climate for students of color and matters for who benefits most from a positive school climate.

Similar to the overall results, we find that student perceptions of both school racial climate and bullying are related to lower likelihood of ODR and suspension, albeit with weaker relationships compared to our peer student behavior and school environment constructs. We find similar differential relationships (i.e., a stronger relationship between disciplinary outcomes and school climate for female students than for male students, for non-SPED students than for SPED students, and for students not from a low socioeconomic background than for students from a low socioeconomic background), with the exception that for school racial climate there is a less strong differential relationship between climate and disciplinary outcomes for Black and White students (i.e., White students still seem to benefit most from better racial school climates, but to a lesser degree compared to peer student behavior or school environment). For our moderation models, we find that some school characteristics (school racial composition and teacher diversity, but not school achievement) still appear to moderate the relationship between school racial climate and discipline outcomes. Though there are many similarities, the differences in results between school climate and racial school climate coupled with prior studies that have highlighted that students of color have lower perceptions of school climate (Anyon et al., 2016; Voight et al., 2015) add support for policymakers and researchers to consider both school climate and school racial climate (Golden et al., 2024; Griffin et al., 2024), especially when examining school discipline.

Policy Implications and Directions for Future Research

First, the study supports the policy recommendation that school climate improves the disciplinary outcomes of students. A growing number of studies have demonstrated the potential of school climate to reduce racial inequality in exclusionary discipline (Bradshaw et al., 2009; Gage et al., 2016; Mitchell et al., 2010). The results underline the importance of school climate for students’ disciplinary outcomes but also highlight key considerations about the theory of action for how improvements in school climate will reduce racial disparities in exclusionary discipline. The benefits of school climate are contextual, and the results illustrate the potential of school climate to reduce racial disparities in school discipline in certain schooling contexts. However, it is possible that system-wide improvements in school climate may reduce exclusionary discipline while widening relative discipline disparities. As such, educational researchers and policymakers should pay greater attention to the potential differential association between school climate and students’ disciplinary outcomes, especially in light of recent studies that have highlighted that school discipline policy and program changes do not seem to improve the disciplinary experiences and outcomes of Black students and alternative approaches to exclusionary discipline are benefiting White students more than Black students (Cruz et al., 2021; Nishioka et al., 2021; Welsh, 2023). If the disciplinary experiences of Black, male, and low-income students do not differentially improve in more positive climates compared to their peers, then how investments in school climate will reduce racial inequities in ODRs and suspensions remains unclear. Encouragingly, this study shows that in schools that are academically low-achieving and schools that serve predominantly students of color, improvements in climate may still differentially benefit Black, Latinx, male, low-income, and special education students, thereby reducing (or at least not increasing) discipline disparities, for both the relative and absolute measures of disparities.

Second, in order to disrupt discipline disparities, the findings suggest that in addition to investing in improving school climate, policymakers may also consider complementary investments in the diversity of school personnel. The study provides novel findings on how the diversity and qualifications of school personnel influences which students most benefit from a positive school climate and adds to the mounting evidence for the benefits of teacher diversity and investments in diversification of the teaching and school leader workforce. The findings illustrate the nuances and interrelation between two common school discipline interventions—improving school climate and diversifying school personnel. Who benefits from school climate matters for school discipline and the personnel in schools moderates who benefits from school climate. The diversity of teachers, principals and assistant principals moderate the extent to which Black students’ disciplinary experiences and outcomes benefit from a more positive school climate (e.g., Plot A in Figure 4). Thus, as policymakers make investments in improving school climate, it is imperative they consider concurrent investment in the diversity of school personnel that influences who benefits most from school climate.

Finally, the paper also offers researchers areas for developing a richer understanding of how school climate influences racial inequality in school discipline. There are still many questions on school climate and exclusionary discipline that abound, such as whether benefits accrue over time. There is a need for qualitative studies that can provide a richer understanding of the relationship between school climate and the likelihood of receiving ODRs and suspensions. Future studies should unpack why school climate matters more in some school contexts than in others.

Supplemental Material

sj-docx-1-epx-10.1177_08959048251340878 – Supplemental material for Examining the Differential Relationships Between School Climate and Students’ Disciplinary Outcomes: Evidence from New York City

Supplemental material, sj-docx-1-epx-10.1177_08959048251340878 for Examining the Differential Relationships Between School Climate and Students’ Disciplinary Outcomes: Evidence from New York City by Richard O. Welsh, Blaise Joseph and Luis A. Rodriguez in Educational Policy