Exclusionary Discipline of Students With Disabilities

Abstract

Given the negative outcomes associated with suspension, scholars and practitioners are concerned with discipline disparities. This study explored patterns and predictors of suspension in a sample of 2,750 students with disabilities in 39 schools in a Midwestern district. Hierarchical generalized linear modeling demonstrated that disability type, gender, race/ethnicity, and free/reduced lunch status were significant predictors of suspension among students with disabilities. Adjusting for gender and race/ethnicity attenuated suspension risk associated with disability type, and adjusting for student-level socioeconomic variables attenuated risk associated with race/ethnicity, but significant disparities remained. School characteristics were not predictive of suspension risk, but their inclusion in the models was associated with increased risk of suspension among students with emotional disturbance. Results underscore the value of multilevel modeling when identifying predictors of suspension and the need to explore a wider variety of classroom and school factors that may account for inequitable discipline.

Keywords

discipline suspension disproportionality students with disabilities multilevel modeling

Ensuring equal access to public education for students with disabilities is one of the primary purposes of the Individuals with Disabilities Education Act (IDEA; 2004). Yet, many schools respond to behavior infractions by suspending students with disabilities, thus removing them from the educational environment, leading legal scholars to suggest that disciplinary exclusion represents de facto denial of educational access (Pauken & Daniel, 2000). Because exclusionary discipline may infringe on students’ rights to a free and appropriate public education, it is imperative that researchers and practitioners examine whether such practices are discriminatory. To this end, scholars often first look for descriptive evidence of group-based differences, known as disparate impact, in outcomes, such as out-of-school suspension. Such evidence serves as a precursor for theories about and investigations into disparate treatment (i.e., discriminatory practices) underlying observed outcomes. Despite decades of research into special education disparities, we are still in the initial stages of understanding the disparate impact of disciplinary exclusion on students with disabilities, which is the focus here.

Prior researchers indicated that exclusionary discipline is disproportionately applied to students with disabilities relative to students who do not have disabilities. Nationally, approximately 7% of students are suspended (Losen & Gillespie, 2012), but estimates for students with disabilities are above 15% and upward of 44% for students identified under the IDEA category of emotional disturbance (ED; Achilles, McLaughlin, & Croninger, 2007). Furthermore, out-of-school suspension of students with disabilities has increased over time (Krezmien, Leone, & Achilles, 2006; Zhang, Katsiyannis, & Herbst, 2004). Among secondary students with ED, in particular, rates of suspension have risen nearly 50% since the 1980s (Wagner, Newman, & Cameto, 2004).

This basic evidence of disparate impact is alarming because suspension is ineffective for reducing inappropriate behavior (Hemphill, Toumbourou, Herrenkohl, McMorris, & Catalano, 2006) and is associated with a variety of negative educational and social outcomes including future disciplinary infractions, repeated suspension, academic failure, school disengagement, and dropout (Arcia, 2006; Scott, Nelson, & Liaupsin, 2001; Skiba & Noam, 2001). Furthermore, by definition, it causes students to miss instruction and opportunities to gain academic and social skills (Christle, Jolivette, & Nelson, 2005), which may exacerbate achievement gaps (Gregory, Skiba, & Noguera, 2010). Together with academic failure and high school dropout—two educational outcomes disproportionately common among students with disabilities—suspension is considered a key contributor to later incarceration of youth and young adults (Christle et al., 2005). Not surprisingly, some have suggested that the overuse of exclusionary discipline with students with disabilities contributes to their overrepresentation in youth detention (Kim, Losen, & Hewitt, 2010).

Demonstrating concern about the potential over-usage of exclusionary discipline with students with disabilities, in 1997, Congress amended IDEA to require states to monitor disparities in long-term suspension and expulsion of students with disabilities and to identify policies, procedures, and practices that may contribute to disproportionate exclusion (IDEA, 1997). More than 15 years later, disparate treatment persists and we still do not understand the dynamics underpinning it. This study attempted to address this gap in the literature by exploring patterns and predictors of out-of-school suspension among students with disabilities.

Predictors of Exclusionary Discipline

Prior researchers identified several sociodemographic characteristics related to differential risk of suspension including: race/ethnicity (Fantuzzo & Perlman, 2007; Vincent, Sprague, & Tobin, 2012), socioeconomic status (SES; Skiba et al., 2011), gender (Pas, Bradshaw, Hershfeldt, & Leaf, 2010), history of previous suspensions (Theriot, Craun, & Dupper, 2010), and special education status (Achilles et al., 2007; Zhang et al., 2004). Scholars have documented the higher risk of suspension among students with disabilities relative to their nondisabled peers (e.g., Sullivan & Bal, 2013; Vincent et al., 2012; Wagner et al., 2004), but few studies have considered potential within-group disparities in suspension among students with disabilities (i.e., differences between disability groups). One exception is Achilles and colleagues (2007) who examined predictors of suspension in a sample of participants from the Special Education Elementary Longitudinal Study (SEELS) who had ED, other health impairments (OHIs), or specific learning disabilities (SLDs). Risk of disciplinary exclusion was highest among students who had ED or OHI, were Black, older, male, had low SES, or attended urban schools. Children who experienced multiple school changes or had parents who expressed low satisfaction with their schools also had higher risk of suspension or expulsion. More recently, Bowman-Perrott and colleagues (2013) analyzed the longitudinal SEELS data to show that disciplinary exclusion was highest among students with ED who were also likely to be excluded multiple times throughout elementary school. Males, Black students, and students with low social skills were also more likely to be suspended over time.

In general, previous work identified disparate impact across various social groupings. This focus on understanding how sociodemographic characteristics relate to suspension risk—as opposed to student behavior (Skiba, 2002)—has been bolstered by decades of research demonstrating Black students receive harsher punishment than their White peers for similar behaviors (e.g., Carter & Jackson, 1982; Skiba, Michael, Nardo, & Peterson, 2002) or more minor offenses (Skiba et al., 2011). Such evidence refutes the claim that racial disparities in suspension simply reflect group differences in behavior.

These findings of both disparate impact and disparate treatment have also led researchers to call for more attention to the relations of characteristics of the school environment to students’ risk of problem behavior (Christle et al., 2005) and suspension (Gregory et al., 2010). School characteristics related to suspension in the general population include percentage of students eligible for free and reduced lunch (Christle, Nelson, & Jolivette, 2004), students’ mobility rate (Mendez, Knoff, & Ferron, 2002), overall level of school academic achievement (Christle et al., 2004), and staff experience (Arcia, 2007). This research suggests that the structural characteristics of schools are related to students’ likelihood of suspension. These studies have rarely considered disability-related disparities in school discipline; however, so we do not know if similar patterns are observable among students with disabilities or comparable with those in general education samples.

Despite the value of this previous research, scholars have generally relied on methods that were unable to capture the complex relationships among multiple predictors or the nested nature of students’ educational experiences. For instance, studies that examined student-level predictors within a single district (e.g., Fantuzzo & Perlman, 2007; Goran & Gage, 2011; Gregory & Weinstein, 2008; Hinojosa, 2008) or school-level characteristics within a district (e.g., Arcia, 2007; Bruns, Moore, Stephan, Pruitt, & Weist, 2005; Christle et al., 2004; Mendez et al., 2002), but not both in tandem, failed to account for the clustering of students in classrooms and schools. Even in studies where both levels of data were available, researchers may not have utilized analyses that accounted for clustering of students in schools because of their reliance on non-parametric tests, correlation, and linear regression (e.g., Theriot et al., 2010; Wu, Pink, Crain, & Moles, 1982). Other studies used data aggregated at the school or district levels to make inferences about individuals, which risk the ecological fallacy by using group-level patterns to infer individual outcomes (Piantadosi, Byar, & Green, 1988). Furthermore, most studies that included school-level factors related to suspension have not considered variables related to special education.

To date, only two published studies have examined predictors of disciplinary exclusion via multilevel analysis. Theriot and colleagues (2010) examined student and school factors predicting whether middle and secondary students from a southeastern school district were suspended for their final discipline infraction of an academic year. They found that poverty, previous behavioral infractions, and severity of the behavior were most predictive of suspension for final infractions. This study is valuable because it utilized multilevel modeling to examine discipline outcomes, but it is limited because the outcome variable reflected only a single discipline infraction. Sullivan, Klingbeil, and Van Norman (2013) examined risk of suspension over an entire academic year in a multilevel sample from an urban school district and found that gender, race, disability, and SES were significant predictors of suspension but that school demographic and performance variables were not. In both cases, these studies included only general disability status (i.e., whether or not a student was identified for special education), not the primary disability, of the students. Thus, more research is needed to tease apart the influence of student characteristics, including specific disability, and school factors in discipline disparities among students with disabilities.

Purpose and Research Questions

In this study, we explored patterns and predictors of suspension among students with disabilities using descriptive analyses and multilevel modeling. We extended research on suspension of students with disabilities in three ways. First, we examined risk of suspension by disability type and sociodemographic characteristics (e.g., SES, gender, race/ethnicity). Numerous scholars have documented racial disparities in suspension, but less attention has been paid to such disparities among children with disabilities (for an exception, see Vincent et al., 2012). Because researchers have demonstrated that students with disabilities generally have an increased risk of suspension relative to students without disabilities (e.g., Sullivan et al., 2013), we limited this study to students with disabilities to explore within-group differences in suspension risk by disability type and other sociodemographics.

Second, by using hierarchical generalized linear models (HGLMs), we accounted for the nesting of students in schools. Most of the previous research has not accounted for the nested nature of students’ school experiences and few discipline studies including students with disabilities examined student and school variables simultaneously. Indeed, only urbanicity has been examined in studies in which special education status or disability was also considered (Achilles et al., 2007; Bowman-Perrott et al., 2013). Thus, we addressed this gap by examining the relations of a variety of school characteristics to suspension among students with disabilities.

Specifically, we explored three models of student-level predictors in the nested models. In the first model, we considered only the relations between suspension and disability. In the second model, we adjusted for age, gender, race/ethnicity, and language status to determine how these demographic characteristics attenuated the relations between disability and suspensions. In the third model, we examined the previous predictors adjusting for variables reflecting SES to ascertain how these variables moderated students’ risk. These latter models were used to evaluate the supposition that differential rates of suspension are due more to socioeconomic factors than race/ethnicity (National Association of Secondary School Principals, 2000; Theriot et al., 2010) or disability—as students of color and those with disabilities are disproportionately poor (Donovan & Cross, 2002).

Third, this study adjusted for specific school-level characteristics to attempt to explain variation in the average probability of suspension for a student receiving special education services in two multilevel models. The first incorporated variables reflecting school size (total enrollment and student–teacher ratio) and sociodemographics (minority, English language learner [ELL], free/reduced lunch enrollment, teacher race, and training) to evaluate the assumption that community sociodemographics, particularly poverty (Achilles et al., 2007), contributed to students’ suspension risk. The final model added variables related to student performance (e.g., reading and mathematics performance, truancy) and policy implementation (retention, special education program size) because previous research has shown that such factors are associated with higher suspension rates (Christle et al., 2005; Flannery, 1997; Hellman & Beaton, 1986; Krezmien et al., 2006). Relations to individual risk among students with disabilities are relatively unexplored.

Thus, our aim was to describe the relations of students’ sociodemographic characteristics and the structural characteristics of their schools to risk of suspension to describe the patterns of disparate discipline among students with disabilities. Our analyses were guided by the following research questions:

Research Question 1: How does risk of suspension vary by disability?

Research Question 2: How do students’ sociodemographics moderate risk of suspension?

Research Question 3: (c) Do school characteristics influence suspension risk among students with disabilities?

Method

Sample

We utilized archival data from a diverse urban school district in the Midwest. After obtaining agreement between the school district and the first author and institutional review board (IRB) approval, the school district provided de-identified student data. During the 2009-2010 school year, the district served 24,295 students in 51 schools. School-level data were obtained from the state’s website (Wisconsin Department of Public Instruction, 2011). Definitions and statutes relating to the data can be obtained from the Wisconsin Department of Public Instruction at www.dpi.wi.gov

The analytic sample included all students with disabilities (N = 2,750) enrolled in 39 schools for which there were complete data. The student-level data had no missing values. Available school-level data for the analytic and full district sample were highly similar with mean estimates for all predictors within ±.10. The correspondence in school-level characteristics provides some evidence that the exclusion of the 12 schools with incomplete data may not bias results. Descriptive statistics for student- and school-level variables for this subset of the analytic sample analyzed are presented in Tables 1 and 2, respectively. The mean age of the sample was 11.28 years.

Table 1.

Descriptive Statistics of the Analytic Sample.

Characteristic	Total	SLD	ID	ED	SLI	OHI	LI
n	2,750	744	187	408	612	404	399
Male	65.93	58.74	54.70	71.08	65.85	70.94	74.19
Race/ethnicity
White	41.93	34.68	36.46	37.75	43.30	43.35	58.65
Black	39.31	48.39	44.75	53.43	27.94	41.87	20.30
Hispanic	12.25	11.02	9.94	6.37	19.93	11.08	11.03
Asian	5.64	5.38	7.73	0.98	8.33	2.71	8.77
Native American	0.87	0.54	1.10	1.47	0.49	0.99	1.25
Limited English proficient	12.76	13.04	13.81	3.19	20.59	9.36	13.03
Free/reduce lunch	64.18	73.39	74.59	76.23	56.54	62.32	43.61
Parent education
Less than HS	10.62	13.31	14.36	10.54	9.31	10.59	6.02
Completed HS	29.24	33.87	38.12	32.84	27.12	28.57	16.79
Some college	26.36	26.34	26.52	34.31	22.22	27.34	23.56
College degree	17.82	14.78	12.71	13.24	19.12	16.50	29.82
Grad/professional degree	15.96	11.69	8.29	9.07	22.22	17.00	23.81

Note. Column percentages are reported. SLD = specific learning disability; ID = intellectual disability; ED = emotional disturbance; SLI = speech-language impairment; OHI = other health impairment; LI = low-incidence disabilities; HS = high school.

Table 2.

Descriptive Statistics for the School-Level Variables (j = 39).

Variable	M	SD	Minimum	Maximum
Enrollment	460.00	299.00	110.00	1,615.00
Student–teacher ratio	9.39	1.80	5.80	12.80
% teachers with master’s degrees or higher	65.00	13.00	39.00	100
% minority enrollment	51.53	13.82	22.70	72.40
% limited English proficient	18.73	10.69	3.60	44.90
% receiving free/reduced lunch	50.22	18.16	21.60	83.10
% White teacher	90.46	5.72	75.00	98.00
% students receiving special education	15.58	4.21	7.00	24.30
% students passing state mathematics test	66.59	12.76	38.80	97.00
% students passing state reading test	72.94	15.12	15.33	72.94
% students retained	0.48	1.10	0.00	6.32
% students suspended	7.06	7.00	0.73	25.45
% students truant	5.52	6.27	0.00	32.70
Drug/weapon incidents per 1,000 students	8.50	12.85	0.00	47.70
Non-drug/weapon incidents per 1,000 students	121.06	134.13	0.00	483.40

Measures

Dependent variable

The dependent measure was out-of-school suspension. Across the entire sample the median number of suspensions per student was 0 (SD = 1.18, range = 0-11). Although multiple suspensions were not uncommon, particularly among students with ED (see Figure 1), we did not treat suspension as a continuous or multinomial outcome in the HGLM for two reasons. First, preliminary analyses demonstrated that the distribution of suspension frequency was highly skewed, violating multiple assumptions required for multilevel modeling with continuous outcomes. Second, additional analyses indicated that given the low number of multiple suspensions relative to the number of students across the sample, models treating suspensions as a multinomial or continuous outcome would likely possess inadequate power to detect statistically significant effects (Raudenbush et al., 2011). Therefore, given the distribution of suspensions and sample size, we treated suspensions as a dichotomous variable in the HGLM. For each student, we created a variable that indicated whether that student was suspended one or more times (1) or not suspended at all (0) during the entire school year. For the descriptive analyses, we created a categorical measure for whether the student was suspended 0, 1, or 2 or more times in the year.

Figure 1.

The proportion of students suspended 0, 1, or ≥2 times in the academic year by disability group.

Student-level predictors

The primary characteristic of interest was the special education disability category of the student, represented by six dummy-coded variables indicating whether a student had a primary disability of SLD, ED, OHI, intellectual disability (ID), ED, speech-language impairment (SLI), or a low-incidence disability (LI). The LI category included all other federally defined disability categories (i.e., autism, deaf-blindness, developmental disability, hearing impairments, multiple disabilities, orthopedic impairments, traumatic brain injury, and visual impairments) that were combined because of low cell size. Because the suspension rate among students with SLI or LI closely approximates the national rate of 7.4% in the general population (Losen & Gillespie, 2012), these categories were combined to serve as the referent group in the HGLM.

Other student-level predictors served as control variables. Race/ethnicity was dummy coded as five variables (Asian/Pacific Islander, Black, Hispanic, Native American, and White). As described in the results below, the majority of suspended students (68%) were Black. As a result, to control for small cell sizes in model fitting, student race/ethnicity was entered as Black or not Black in the HGLM. Sensitivity analyses supported this decision from an analytic standpoint.

Additionally, student gender, language status (whether the student was classified as has having limited English proficiency [LEP]), eligibility for free or reduced lunch, and a series of dummy-coded variables for the highest parental education level were reported. Finally, we included a continuous variable for age of the student.

School-level predictors

The following school-level predictors, which were reported as building-level percentages, served as additional control variables: minority enrollment, LEP enrollment, special education eligibility, free/reduced lunch eligibility, truancy, students retained, students meeting state standards in reading and math state achievement tests, teachers with master’s degrees or higher, and teachers who were White. In addition, we included total enrollment, student–teacher ratio, and incident rates per 1,000 students for drug/weapon offenses and all other offenses.

Analyses

Descriptive analyses

We estimated race-based disparities in suspension by calculating the percentage of total suspensions comprised by each racial/ethnic group; percentage of students within each racial group suspended (i.e., risk of suspension); and relative risk of racial/ethnic minority students’ suspension compared with White students. Relative risk is calculated as a ratio of two groups’ risk (e.g., Black students’ risk of suspension divided by White students’ risk).

Multilevel analysis

To determine whether multilevel modeling was appropriate, that is, to determine whether the likelihood of being suspended was related to the school in which a student was enrolled and not to student-level characteristics alone, an unconditional model was first fitted with no level 1 or 2 predictors ( ${\hat{γ}}_{00} = - 1.74$ ). The model-implied probability of a student in special education being suspended one or more times in a school with a typical rate of suspensions for special education students (µ_0j = 0) was equal to .15. A significant chi-square test statistic for the variation of intercepts, τ₀₀ = .81 (χ² = 439.37, p < .001) suggested that multilevel modeling was appropriate. The pseudo intraclass correlation coefficient (ICC; Snijders & Bosker, 1999) for the empty model was 0.04 (see Note 1).

Next, we fit five successive models using HLM 7 (Raudenbush, Bryk, Cheong, Congdon, & Toit, 2011). The first three models evaluated Level 1 (student-level) predictors. The remaining two models added Level 2 (school-level) predictors to attempt to model variation in the intercept. We grand-mean centered age at Level 1, as well as all predictors at Level 2, to facilitate meaningful interpretation of intercept estimates. To simplify model building and interpretation, slopes were held constant for each Level 1 covariate. We attempted to allow slopes to vary after initial model building but doing so resulted in multiple convergence errors during estimation.

Link function

Given the dichotomous nature of the outcome variable (0 = no suspensions, 1 = one or more suspensions), assumptions for using typical hierarchical analysis could not be satisfied, particularly normality and heteroscedasticity of Level 1 residuals (Raudenbush & Bryk, 2002). Our primary interest was predicting the likelihood of suspension conditioned on student- and school-level predictors. To accomplish this, it is typical to perform a nonlinear transformation of the Level 1 predicted value using a link function. We used the following logit link function:

{\hat{η}}_{i j} = \ln (\frac{{\hat{ρ}}_{i j}}{1 - {\hat{ρ}}_{i j}}),

where ${\hat{η}}_{i j}$ is the predicted log odds of suspension of a student and ${\hat{ρ}}_{i j}$ is the model-implied probability of suspension. If the probability of success of a student, expressed as ψ_ij, is < .5, then ${\hat{η}}_{i j}$ will be negative; if ψ_ij is >.5, then ${\hat{η}}_{i j}$ will be positive. As a result, a one-unit increase in a predictor increases the log odds of suspension of the predictor variable. Transforming or “linearizing” the outcome variable satisfied the conditions for multilevel modeling. The predicted log odds can be re-expressed as a probability by returning the value to a logit expression. A Bonferroni correction was used to control for chance significance as a result of conducting multiple comparisons. Thus, for each model fit, the Type I error rate was equal to .15/n, where n is the number of fixed effects being estimated.

Results

Descriptive Analysis

The descriptive analyses of risk and relative risk are shown in Table 3. In the analytic sample, 19.5% of students with disabilities were suspended at least once. Risk of suspension varied substantially by race/ethnicity. Black students were nearly 3 times more likely to be suspended than White students, whereas Hispanic and Asian students were 20% and 66% less likely to be suspended, respectively.

Table 3.

Number and Percentage of Students in Special Education Enrollment and Suspension by Race/Ethnicity: Risk and Relative Risk of Suspension.

	Enrolled		Suspended
Race/ethnicity	n	%	n	%	Risk	Relative risk^a
Black	1,081	39.31	362	67.54	0.33	2.90
Hispanic	337	12.25	31	5.78	0.09	0.80
Asian	155	5.64	6	1.12	0.04	0.34
Native American	24	0.87	4	0.75	0.17	1.44^b
White	1,153	42.93	133	24.81	0.12	^a
Total	2,750	100	536	100	0.19	—

White students were the referent group.

Interpret with caution given small cell sizes.

The differences in suspension risk by disability category are shown in Figure 1. Among all students with disabilities, 8.8% were suspended once and 10.7% were suspended twice or more. Suspension rates were highest for students with ED and lowest for students with SLI. Nearly half of the students with ED were suspended at least once; 30% were suspended multiple times even though multiple suspensions were relatively infrequent in the other disability categories, particularly SLI and LI. Overall risk of suspension was fairly consistent for students identified as SLD, ID, and OHI, with 18% to 22% of these students suspended at least once.

Multilevel Modeling

We evaluated five models of suspension risk using hierarchical logistic regression. Model 1 included only disability category. Models 2 and 3 incorporated sociodemographic characteristics. Model 4 added school enrollment and demographic characteristics of teachers and students. The final model, Model 5, added variables related to student performance and policy implementation. The results are summarized in Table 4.

Table 4.

Multilevel Logistic Regression Results.

	Model 1			Model 2			Model 3			Model 4			Model 5
Variable	β	SE _β	e ^β	β	SE _β	e ^β	β	SE _β	e ^β	β	SE _β	e ^β	β	SE _β	e ^β
Intercept	−2.85^†	0.20	0.06	−3.74*	0.25	0.02	−4.18**	0.32	0.02	−4.35***	0.33	0.01	−4.49****	0.29	0.01
Level 1
SLD	0.93^†	0.18	2.54	0.68*	0.17	1.98	0.54**	0.18	1.71	0.55***	0.18	1.73	0.55****	0.18	1.73
ID	0.76^†	0.24	2.15	0.59	0.26	1.81	0.39	0.28	1.48	0.41	0.28	1.50	0.42	0.28	1.53
ED	2.24^†	0.22	9.39	1.97*	0.18	7.17	1.87**	0.20	6.49	1.89***	0.20	6.62	1.91****	0.21	6.78
OHI	1.15^†	0.19	3.16	0.92*	0.21	2.52	0.87**	0.21	2.40	0.87***	0.22	2.40	0.89****	0.22	2.43
Age				0.01	0.06	1.02	0.04	0.05	1.04	0.00	0.08	1.00	−0.05	0.08	0.95
Gender				0.66*	0.18	1.93	0.68**	0.18	1.96	0.68***	0.18	1.98	0.70****	0.18	2.01
Black				1.28*	0.13	3.60	0.87**	0.13	2.38	0.86***	0.13	2.36	0.86****	0.14	2.37
LEP				−0.26	0.23	0.77	−0.63**	0.22	0.53	−0.62***	0.22	0.53	−0.63	0.22	0.53
FRL							0.93**	0.14	2.55	0.93***	0.14	2.55	0.93****	0.14	2.52
Parent education—HS							0.43**	0.13	1.54	0.44***	0.13	1.56	0.47****	0.13	1.60
Parent education—some college							−0.19	0.19	0.83	−0.17	0.19	0.84	−0.15	0.19	0.86
Parent education—college							−0.18	0.21	0.83	−0.16	0.22	0.86	−0.15	0.22	0.86
Parent education—graduate							0.07	0.22	1.07	0.11	0.22	1.11	0.13	0.22	1.14
Level 2
Enrollment										0.00	0.00	1.00	0.00	0.00	1.00
Student–teacher ratio										0.14	0.13	1.15	0.04	0.17	1.04
Teachers with master’s degree										−0.13	1.06	0.87	1.10	0.70	2.99
% minority										0.02	0.03	1.02	−0.02	0.02	0.98
% LEP										−0.04	0.02	0.96	−0.00	0.01	1.00
% White teachers										−3.38	2.96	0.03	−0.52	2.17	0.59
% FRL										0.01	0.02	1.01	−0.00	0.01	1.00
% SPED													−0.03	0.01	0.97
% retained													−0.31	0.27	0.74
% pass math													−0.03	0.05	0.97
% pass reading													0.00	0.01	1.00
% Truant													0.02	0.05	1.02
Suspension													0.02	0.03	1.02
Incidents—drug/weapon													0.01	0.01	1.01
Incidents—non-drug/weapon													0.01****	0.00	1.00

Note. e^β = log odds; SLD = specific learning disability; ID = intellectual disability; ED = emotional disturbance; SLI = speech-language impairment; OHI = other health impairment; LEP = Limited English proficiency; FRL = free/reduced lunch; HS = high school; SPED = special education.

†

p < .03 (.15/5). *p < .02 (.15/9). **p < .011 (.15/14). ***p < .007 (.15/21). ****p < .005 (.15/29).

Disability status

For Model 1, statistically controlling for other disability categories, having any special education category other than SLI or LI increased the likelihood of a student being suspended. The ED category showed the most pronounced effect, as these students were 9 times as likely to be suspended as students with SLI or LI.

Race/ethnicity, age, gender, and language

Gender and race/ethnicity had statistically significant effects controlling for all other predictors on the likelihood of suspension. In particular, being Black was associated with a 3.6 times greater risk of suspension. In addition, adding these predictors attenuated the effect of disability category.

SES

Students were more likely to be suspended if they received free/reduced lunch or if their parents had only a high school diploma or less education. These predictors accounted for some of the variation in suspension likelihood and reduced the odds of suspension associated with disability and race/ethnicity but not with age or gender.

School enrollment and teacher characteristics

Model 4 added covariates related to student enrollment, teacher characteristics (e.g., race/ethnicity, highest degree) and student-to-teacher ratio. None of these predictors were significant at the corrected Type I error rate. Furthermore, they did not diminish the relations between student-level characteristics and suspension risk.

Student performance and policy proxies

Like the previous model, these school characteristics generally did not predict suspension risk and did not attenuate the relations of risk and student characteristics. Instead, accounting for school characteristics resulted in an increase in the log odds of suspension among students with ED. The only significant predictor was the rate of non-drug/weapon-related disciplinary infractions. The small coefficient indicated that as the rate of non-drug/weapon-related disciplinary infractions increased at the school level, there was a small increase in the average log odds of suspension for a student receiving special education, statistically controlling for other predictors.

Discussion

We examined patterns and predictors of suspension among students with disabilities using descriptive analyses and multilevel modeling to identify the student and school sociodemographic and structural characteristics related to discipline disparities. The results indicate variability in suspension risk across disability categories, highlighting the need to disaggregate samples of students with disabilities when analyzing discipline disparities. Nationally, 7.4% of all students are suspended each year (Losen & Gillespie, 2012); however, 19% of our sample of students with disabilities was suspended at least once in an academic year, with the disability groups varying from 7% to 47%. Multiple suspensions were commonplace, particularly among students with ED or OHI. The results of the HGLM also indicated significant risk associated with race and gender, echoing earlier studies demonstrating racial/ethnic and gender disparities in discipline of students with disabilities (e.g., Bowman-Perrott et al., 2013; Vincent et al., 2012).

These results suggest suspension is particularly problematic among students with ED. The overall suspension rate for this group (47%) was commensurate with nationally representative estimates (Wagner et al., 2004), but nearly one third of the students with ED received multiple suspensions, nearly twice the proportion that received a single suspension, suggesting that this disciplinary consequence is both disproportionately applied and ineffective for these students. Given the relations of exclusionary discipline to poor academic outcomes, delinquency, disengagement, and dropout, the potential overuse of suspension with this group may exacerbate their behavioral and social–emotional problems and poor educational outcomes. In addition, OHI was also positively related to suspension risk in the regression analysis when other categories and characteristics were controlled for, consistent with earlier research documenting high rates of suspension among students with OHI due to attention deficit hyperactivity disorder (ADHD; Achilles et al., 2007). The disproportionate suspension of students with ED and OHI calls for further attention in research and practice.

A key finding of this study was that the school variables examined in the multilevel analyses did not predict suspension. Many of these constructs have shown significant relations to discipline outcomes in analyses of building-level suspension rates (e.g., Bruns et al., 2005; Christle et al., 2004; Mendez et al., 2002; Wu et al., 1982). However, such studies generally relied on descriptive analyses, correlations, or school/district aggregates that did not allow for estimation for the relations to individuals’ risk of suspension. The differences in approaches and the resultant inferences permitted by these analyses may account for the discrepancy between the prior and current findings for the relations of these school characteristics to suspension risk. Although the present results do not discount the importance of school characteristics in suspension risk, they suggest that the school characteristics examined may not be of central importance. Instead, scholars should examine other features of the educational environment to understand better how schools influence discipline outcomes (e.g., school policies and procedures, teacher and administrator perceptions and practices). In summary, the multilevel modeling provided support for previous findings related to sociodemographic characteristics and risk, but the results also challenge earlier findings regarding the relations of school factors in studies where researchers relied on bivariate, single-level analysis, and/or building aggregates. These findings should be replicated in other samples to ascertain how they apply across contexts. Further research is needed to explore the causes and consequences of differential risk.

Limitations and Future Research

Although this study extends the current research, we must note a few limitations. First, an important limitation of the present study was the use of secondary data analysis on a single-school system, which limited constructs considered and generalizability of the findings. Typical model diagnostic statistics could not be calculated. Thus, the results should be interpreted as exploratory analysis of the potential effects of different individual- and school-level predictors. The lack of significant school-level predictors should not be taken to indicate the lack of a relationship between suspensions and school context. The data available in this data set may not have been appropriate to capture the complexity of the school factors influencing disproportionate discipline. It would be particularly beneficial to analyze large-scale samples to replicate and extend these findings. In addition, there is considerable value in exploring the relations of students’ behavioral and disciplinary outcomes to classroom and school factors related to behavior management—individual- and systems-level implementation of positive behavioral interventions and supports (PBIS), for instance—to elucidate how such practices may contribute to or reduce disparate impact and treatment (Sullivan et al., 2013).

Finally, we could not disentangle behavior, referral, and suspension in the present analysis, although we did select the child characteristics that should be the focus of future research. Research is needed to discern whether students with disabilities, and specific subgroups, are more likely to engage in serious behavioral infractions, or whether disciplinary consequences are disproportionately applied to certain groups. Researchers should also examine the context of suspension (e.g., behavior, referral source, time). Scholars emphasized that differences in behavioral frequency alone simply do not account for racial disparities in discipline (e.g., Skiba et al., 2011), and that, instead, it is likely differential selection in referral and administrative decisions account for disparities in suspension (Gregory et al., 2010). This differential selection hypothesis also may apply to disparities in the use of exclusionary discipline with students with disabilities, as teachers and administrators may systematically respond differently to the behaviors of students with special needs, particularly those with ED.

Several existing lines of scholarship can be brought together in efforts to understand disability-related discipline disparities. Scholars often have suggested that the behaviors of students with cultural or physical differences are misinterpreted or pathologized (Cartledge & Kourea, 2008). In this vein, researchers can explore how generalized conceptualizations and contextually grounded interpretations of behavior affect disciplining students with disabilities. Given evidence that (a) race/ethnicity influences teachers’ expectations of students’ capabilities (Cartledge & Kourea, 2008) and (b) Black students are suspended more because of disciplinary referrals for milder, subjective behaviors (e.g., disrespect; Skiba et al., 2002), scholars have called for exploration of the relationships between race/ethnicity and exclusion. The present results suggest the need to consider the complex interplay of race/ethnicity, gender, and disability in exclusion as well. In examining the dynamics underpinning observed disparities to understand why disability and race/ethnicity (both separately and in combination) are associated with higher rates of suspension, special education scholars may consider extending research from social psychology. For instance, early social psychology research showed that ambiguously aggressive behaviors were perceived more negatively when the actor was Black than when the actor was White (Sagar & Schofield, 1980) and that people judged Black juvenile offenders more harshly than White juveniles for identical offenses and assigned more severe punishment (Rattan, Levine, Dweck, & Eberhardt, 2012). Education researchers could test the extent to which race/ethnicity and disability, and various combinations thereof, affect referral and administrative decisions, especially given that many educators are likely unaware of their biases or how their biases affect their behaviors in the classroom (Cartledge & Kourea, 2008). This, in turn, could inform professional learning and policy to foster more equitable behavioral supports for students with disabilities.

Educational Policy and Practice Implications

This analysis calls attention to the need, in both policy and practice, to address the widespread disciplinary exclusion of students with ED, as well as students with disabilities more generally. The present findings bring to light two broader policy issues: (a) whether disproportionality in discipline is due to the severe behavioral needs of these students, or (b) whether the observed disparities indicate inappropriate overreliance on exclusionary discipline (i.e., disparate treatment). Repeated use of exclusionary discipline with individuals with disabilities may indicate a failure of schools to address students’ special needs and denial of procedural protections (Kim et al., 2010). High suspension rates among students with ED are of particular concern given the defining characteristics of the disability category. IDEA calls for use of functional behavioral assessment and research-based strategies, including PBIS, to address students’ behavioral difficulties; the extent to which this is not happening on a large scale—which the present data in conjunction with findings from nationally representative samples would suggest—represents a national and local policy issue. Where race, gender, and disability intersect to amplify students’ risk of exclusion, additional attention is warranted to prevent ongoing disparities.

Given evidence of the negative educational outcomes associated with suspension, care must be taken to minimize these effects and to prevent disparate impact on already vulnerable populations. Educators and administrators should reconsider their reliance on exclusionary discipline and take steps to prevent differential use. Policies pertaining to behavioral expectations and consequences should be examined for potential biases and disparate impact (Townsend, 2000). Policies that foster reliance on exclusionary discipline with students with disabilities contravene IDEA’s support for research-supported strategies (Krezmien et al., 2006) and are unlikely to curb unwanted behaviors. In cases of escape/avoidance-motivated behaviors, suspension may actually reinforce students’ negative behaviors (Skiba et al., 2002). Policy and policy implementation should be reviewed to ensure that identified consequences are appropriate to the behavior and are consistent with broader educational objectives. Schools’ responsibility to ensure appropriate education is central to the spirit and requirements of IDEA. Exclusionary discipline may undermine this obligation (Skiba, 2002) and negate the most basic goals of special education (Pauken & Daniel, 2000).

The possibility that some students are discriminatorily suspended is an ethical dilemma for educators (Cartledge, Tillman, & Johnson, 2001). Practices should be structured to ensure that educators are not inadvertently contributing to increased marginalization of students with disabilities through inappropriate use of exclusionary discipline. Such tactics should be reserved for only the most serious offenses; for all others, PBIS and other alternatives to suspension (e.g., Chin, Dowdy, Jimerson, & Rime, 2012) should be used. This must be underpinned by a shift from a punitive orientation to one that emphasizes fostering appropriate behavior (Cartledge et al., 2001).

Conclusion

Results of the present study add to the evidence that students with disabilities may be overexposed to exclusionary discipline despite the intended protections of the IDEA. Suspension risk varies by disability and other sociodemographic characteristics, highlighting the need for research to examine how discipline policies and practices may contribute to observed disparities. The high rate of suspension among students with disabilities, especially those with ED and OHI, also underscores the need for more proactive behavioral supports for students with disabilities to prevent further marginalization and exacerbation of educational difficulties through forced exclusion.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Achilles

G. M.

McLaughlin

M. J.

Croninger

(2007). Sociocultural correlates of disciplinary exclusion among students with emotional, behavioral, and learning disabilities in the SEELS national dataset. Journal of Emotional and Behavioral Disorders, 15, 33–45. doi:10.1177/10634266070150010401

Arcia

(2006). Achievement and enrollment status of suspended students: Outcomes in a large, multicultural school district. Education and Urban Society, 38, 359–369. doi:10.1177/0013124506286947

Arcia

(2007). Variability in schools’ suspension rates of black students. Journal of Negro Education, 76, 597–608.

Bowman-Perrott

Benz

M. R.

Hsu

Kwok

Eisterhold

Zhang

(2013). Patterns and predictors of disciplinary exclusion over time: An analysis of the SEELS national dataset. Journal of Emotional and Behavioral Disorders, 21, 83–96.

Bruns

E. J.

Moore

Stephan

S. H.

Pruitt

Weist

M. D.

(2005). The impact of school mental health services on out-of-school suspension rates. Journal of Youth and Adolescence, 34, 23–30. doi:10.1007/s10964-005-1333-z

Carter

Jackson

(1982). Student discipline, disproportionality and the law. In Harris

Bennett

(Eds.), Student discipline: Legal, empirical, and educational perspectives (pp. 37–48). Bloomington: Indiana University.

Cartledge

Kourea

(2008). Culturally responsive classrooms for culturally diverse students with and at risk for disabilities. Exceptional Children, 74, 351–371.

Cartledge

Tillman

L. C.

Johnson

C. T.

(2001). Professional ethics within the context of student discipline and diversity. Teacher Education and Special Education, 24, 25–37. doi:10.1177/088840640102400105

Chin

J. K.

Dowdy

Jimerson

S. R.

Rime

(2012). Alternatives to suspension: Rationale and recommendations. Journal of School Violence, 11, 156–173.

10.

Christle

Jolivette

Nelson

M. C.

(2005). Breaking the school to prison pipeline: Identifying school risk and protective factors for youth delinquency. Exceptionality, 13, 69–88. doi:10.1207/s15327035ex1302_2

11.

Christle

Nelson

M. C.

Jolivette

(2004). School characteristics related to the use of suspension. Education & Treatment of Children, 27, 509–526.

12.

Donovan

M. S.

Cross

C. T.

(Eds.). (2002). Minority students in special and gifted education. Washington, DC: National Academies Press.

13.

Fantuzzo

Perlman

(2007). The unique impact of out-of-home placement and the mediating effects of child maltreatment and homelessness on early school success. Children and Youth Services Review, 29, 941–960. doi:10.1016/j.childyouth.2006.11.003

14.

Flannery

D. J.

(1997). School violence: Risk, prevention, intervention, and policy (Report No. RR93002016). Retrieved from http://www.eric.ed.gov/PDFS/ED416272.pdf

15.

Goran

L. G.

Gage

N. A.

(2011). A comparative analysis of language, suspension, and academic performance of students with emotional disturbance and students with learning disabilities. Education & Treatment of Children, 34, 469–488. doi:10.1353/etc.2011.0035

16.

Gregory

Skiba

Noguera

(2010). The achievement gap and the discipline gap: Two sides of the same coin? Educational Researcher, 39, 59–68. doi:10.3102/0013189x09357621

17.

Gregory

Weinstein

(2008). The discipline gap and African Americans: Defiance or cooperation in the high school classroom. Journal of School Psychology, 46, 455–475. doi:10.1016/j.jsp.2007.09.001

18.

Hellman

D. A.

Beaton

(1986). The pattern of violence in urban public schools: The influence of school and community. Journal of Research in Crime & Delinquency, 23, 102–127. doi:10.1177/0022427886023002002

19.

Hemphill

S. A.

Toumbourou

J. W.

Herrenkohl

T. I.

McMorris

B. J.

Catalano

R. F.

(2006). The effect of school suspensions and arrests on subsequent adolescent antisocial behavior in Australia and the United States. Journal of Adolescent Health, 39, 736–744. doi:10.1016/j.jadohealth.2006.05.010

20.

Hinojosa

M. S.

(2008). Black-white differences in school suspension: Effect of student beliefs about teachers. Sociological Spectrum, 28(2), 175–193. doi:10.1080/02732170701796429

21.

Individuals with Disabilities Education Act of 1997, 20 U.S.C. § 1400 et seq. Retrieved from http://www2.ed.gov/policy/speced/leg/idea/idea.pdf

22.

Individuals with Disabilities Education Act of 2004, 20 U.S.C. § 1400 et seq. Retrieved from http://idea.ed.gov/download/statute.html

23.

Kim

C. Y.

Losen

D. J.

Hewitt

(2010). The school-to-prison pipeline: Structuring legal reform. New York: New York University Press.

24.

Krezmien

M. P.

Leone

P. E.

Achilles

G. M.

(2006). Suspension, race, and disability: Analysis of statewide practices and reporting. Journal of Emotional and Behavioral Disorders, 14, 217–226. doi:10.1177/10634266060140040501

25.

Losen

D. J.

Gillespie

(2012). Opportunities suspended: The disparate impact of disciplinary exclusion from school. Los Angeles, CA: The Center for Civil Rights Remedies.

26.

Mendez

L. M.

Knoff

H. M.

Ferron

J. M.

(2002). School demographic variables and out-of-school suspension rates: A quantitative and qualitative analysis of a large ethnically diverse school district. Psychology in the Schools, 39, 259–277. doi:10.1002/pits.10020

27.

National Association of Secondary School Principals. (2000, February). Statement on civil rights implications of zero tolerance programs. Washington, DC: Testimony presented to the U.S. Commission on Civil Rights.

28.

Pas

E. T.

Bradshaw

C. P.

Hershfeldt

P. A.

Leaf

P. J.

(2010). A multilevel exploration of the influence of teacher efficacy and burnout on response to student problem behavior and school-based service use. School Psychology Quarterly, 25, 13–27. doi:10.1037/a0018576

29.

Pauken

P. D.

Daniel

P. T. K.

(2000). Race discrimination and disability discrimination in school discipline: A legal and statistical analysis. West’s Education Law Reporter, 139, 759–790.

30.

Piantadosi

Byar

D. P.

Green

S. B.

(1988). The ecological fallacy. American Journal of Epidemiology, 127, 893–904.

31.

Rattan

Levine

C. S.

Dweck

C. S.

Eberhardt

J. L.

(2012). Race and the fragility of the legal distinction between juveniles and adults. PLoS ONE, 7(5), 1–5.

32.

Raudenbush

S. W. et al.

(2011). Optimal Design Software for Multi-level and Longitudinal Research (Version 3.01) [Software]. Retrieved from http://sitemaker.umich.edu/group-based/optimal_design_software

33.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: SAGE.

34.

Raudenbush

S. W.

Bryk

A. S.

Cheong

Y. F.

Congdon

R. T.

Toit

(2011). HLM7: Hierarchical linear and nonlinear modeling. Lincolnwood, IL: Scientific Software International.

35.

Sagar

H. A.

Schofield

J. W.

(1980). Racial and behavioral cues in black and white children’s perceptions of ambiguously aggressive acts. Journal of Personality and Social Psychology, 39, 590–598. doi:10.1037/0022-3514.39.4.590

36.

Scott

T. M.

Nelson

C. M.

Liaupsin

C. J.

(2001). Effective instruction: The forgotten component in preventing school violence. Education & Treatment of Children, 24, 309–322.

37.

Skiba

R. J.

(2002). Special education and school discipline: A precarious balance. Behavioral Disorders, 27, 81–97.

38.

Skiba

R. J.

Horner

R. H.

Chung

C.-G.

Rausch

May

Tobin

(2011). Race is not neutral: A national investigation of African American and Latino disproportionality in school discipline. School Psychology Review, 40, 85–107.

39.

Skiba

R. J.

Michael

R. S.

Nardo

A. C.

Peterson

R. L.

(2002). The color of discipline: Sources of racial and gender disproportionality in school punishment. The Urban Review, 34, 317–342.

40.

Skiba

R. J.

Noam

G. G.

(Eds.). (2001). Zero tolerance: Can suspension and expulsion keep schools safe? San Francisco, CA: Jossey-Bass.

41.

Snijders

T. A. B.

Bosker

R. J.

(1999). Multilevel analysis: An introduction to basic and advanced multilevel modeling. London: Sage Publications.

42.

Sullivan

A. L.

Bal

(2013). Disproportionality in special education: Effects of individual and school variables on risk. Exceptional Children, 79, 475–494.

43.

Sullivan

A. L.

Klingbeil

Van Norman

(2013). Beyond behavior: Multilevel analysis of the influence of sociodemographics and school characteristics on students’ risk of suspension. School Psychology Review, 47, 99–114.

44.

Theriot

M. T.

Craun

S. W.

Dupper

D. R.

(2010). Multilevel evaluation of factors predicting school exclusion among middle and high school students. Children and Youth Services Review, 32, 13–19. doi:10.1016/j.childyouth.2009.06.009

45.

Townsend

B. L.

(2000). The disproportionate discipline of African American learners: Reducing school suspensions and expulsions. Exceptional Children, 66, 381–391.

46.

Vincent

Sprague

J. R.

Tobin

(2012). Exclusionary discipline practices across students’ racial/ethnic backgrounds and disability status: Findings from the Pacific Northwest. Education & Treatment of Children, 35, 585–601.

47.

Wagner

Newman

Cameto

(2004). Changes over time in the secondary school experiences of students with disabilities: A special topic report of findings from the National Longitudinal Transition Study–2 (NLTS2). Menlo Park, CA: SRI International.

48.

Wisconsin Department of Public Instruction. (2011). Wisconsin Information Network for Successful Schools [online database]. Retrieved from http://dpi.wi.gov/sig/index.html

49.

Pink

Crain

Moles

(1982). Student suspension: A critical appraisal. Urban Review, 14, 245–283.

50.

Zhang

Katsiyannis

Herbst

(2004). Disciplinary exclusions in special education: A four-year analysis. Behavioral Disorders, 29, 337–347.