Contents lists available at ScienceDirect
Journal of School Psychology
journal homepage: www.elsevier.com/locate/jschpsyc
Are students with disabilities suspended more frequently than
otherwise similar students without disabilities?
Paul L. Morgan
, George Farkas
, Marianne M. Hillemeier
, Yangyang Wang
, Christopher DeJarnett
, Steve Maczuga
The Pennsylvania State University, United States of America
University of California, Irvine, United States of America
Action Editor: Amy Briesch
Students with disabilities (SWD) have been reported to be disproportionately suspended from
U.S. schools and so more likely to experience the “school-to-prison pipeline” through suspension's
associations with lower academic achievement, dropout, juvenile delinquency, and adult crim-
inality. Yet few studies have estimated SWD's risk of more frequent suspension while simulta-
neously controlling for potential confounds. Negative binomial regression modeling of suspen-
sion count data from a nationally representative and longitudinal sample (N= 6,740) indicated
that males, those from lower resourced families, and students attending more economically
segregated schools were more frequently suspended. On average, students who are Black re-
ceived about 1.6 times as many suspensions by the end of 8th grade as otherwise similar White
students. In contrast, having a disability by 1st grade was not a risk factor for more frequent
suspension by the end of 8th grade while simultaneously accounting for other risk factors (e.g.,
gender, race/ethnicity, family SES, prior history of externalizing problem behaviors, being from a
English-speaking household, school-level economic composition). Students with speciﬁc dis-
ability conditions (e.g., emotional disturbances, speech or language impairments) were not at
increased risk for more frequent suspension. Students with disabilities who are Black, Hispanic,
or of other race/ethnicity were not more frequently suspended than SWD who are White.
U.S. schools are systems for academically, behaviorally, and socially educating students. Schools are therefore expected to be safe,
orderly, and civil environments where criminal, violent, or threatening behaviors rarely, if ever, occur. Schools may use suspensions
to deter students from engaging in such behaviors. Speciﬁcally, suspension is designed to decrease the likelihood that violent or
seriously disruptive students continue to engage in potentially dangerous behaviors as well as to protect the safety and ensure the
education of other students (Lamont et al., 2013). Seriously disruptive students who remain in classrooms often adversely aﬀect the
learning and behavior of their peers (Carrell & Hoekstra, 2010;Figlio, 2007;Fletcher, 2010;Gottfried, Egalite, & Kirksey, 2016;Horoi
Received 11 April 2017; Received in revised form 22 October 2018; Accepted 28 November 2018
Funding support was provided by Spencer Foundation Midcareer Grant to the ﬁrst author as well as an infrastructure grant (P2CHD041025),
Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health. No oﬃcial endorsement should
Corresponding author at: Department of Education Policy Studies, 310E Rackley Building, The Pennsylvania State University, University Park,
PA 16802, United States of America.
E-mail address: firstname.lastname@example.org (P.L. Morgan).
Journal of School Psychology 72 (2019) 1–13
0022-4405/ © 2018 Published by Elsevier Ltd on behalf of Society for the Study of School Psychology.
& Ost, 2015;Neidell & Waldfogel, 2010). About 7 million students, or about 14% of the school-aged population, were suspended from
U.S. schools in 2011–2012 (U.S. Department of Education, 2016a, 2016b).
Despite its use by U.S. schools, suspension is increasingly questioned as a disciplinary practice because of its possible adverse
eﬀects on students who are suspended. For example, Morris and Perry's (2016) analyses of a longitudinal sample of students in
Kentucky indicated that suspended students tended to later display lower reading and mathematics achievement. Noltemeyer, Ward,
and Mcloughlin's (2015) meta-analysis of 34 studies indicated that suspended students were more likely to drop out of school. Use of
suspension may result in large ﬁscal costs for U.S. states through lost future tax revenue (Rumberger & Losen, 2016). Both cross-
sectional and longitudinal studies indicate that students who are suspended are more likely to engage in substance abuse and violent
behaviors as well as to be referred to courts during adolescence (Hemphill et al., 2009;Hemphill, Heerde, Herrenkohl, Toumbourou,
& Catalano, 2012;Nicholson-Crotty, Birchmeier, & Valentine, 2009). Students who are suspended are also more likely to later engage
in violent criminality and to be arrested in adulthood (Katsiyannis, Thompson, Barrett, & Kingree, 2013;Mowen & Brent, 2016). By
increasing the risk of academic diﬃculties, school dropout, substance abuse, juvenile delinquency, and adult criminality and arrest,
suspension is viewed as an entry point into the metaphorical “school-to-prison pipeline” in which school-based disciplinary practices
and policies increase the risk that students interact with the juvenile and adult incarceration systems (Hemphill et al., 2009;Mowen &
Brent, 2016). For example, being suspended from school is positively associated with committing crimes, including felonies, that
result in arrest and referral to the juvenile justice system (Cuellar & Markowitz, 2015;Mittleman, 2018).
1.1. Disparities in the use of suspension by U.S. schools
In addition to suspension's associations with lower academic achievement, dropout, juvenile delinquency, and adult criminality,
there is also suggestive evidence that U.S. schools may discriminate in their use of suspension. For example, students who are racial or
ethnic minorities have been reported to be more likely to be suspended than students who are White (Skiba et al., 2014;U.S.
Government Accountability Oﬃce [GAO], 2018). These disparities can occur even among otherwise similarly behaving students
(Horner, Fireman, & Wang, 2010;Petras, Masyn, Buckley, Ialongo, & Kellam, 2011), as well as those attending similarly resourced
schools (GAO, 2018). Racial disparities in suspension are thought to at least partially explain achievement gaps (Gregory, Skiba, &
Noguera, 2010;Morris & Perry, 2016). Males, students being raised in low-income households, as well as those who are older for their
grade level have also been found to be more likely to be suspended, including among otherwise similarly behaving students (Petras
et al., 2011;Theriot, Craun, & Dupper, 2010), thereby suggesting that U.S. schools may be discriminating against some socio-
demographic groups in their disciplinary practices (National Research Council [NRC], 2004). However, and alternatively, it may be
that school contexts partially explain these disparities, particularly for racial or ethnic minorities and those from low-income families
who more frequently attend schools where “zero tolerance” disciplinary practices are in use (Kinsler, 2011). Thus, what is attributed
to discrimination due to diﬀerential treatment of similarly situated students may instead result from minority students being more
likely to attend schools whose disciplinary practices involve greater use of suspension.
Students with disabilities (SWD) have also been reported to be disproportionately suspended (GAO, 2018;Sullivan, Klingbeil, &
Van Norman, 2013), leading to suggestions that schools may be discriminating in regards to disability status (Kim, Losen, & Hewitt,
2010;Losen & Gillespie, 2012). For example, suspension rates for students with and without disabilities are estimated to be about
12–15% and 6–7%, respectively (Achilles, McLaughlin, & Croninger, 2007;Losen & Gillespie, 2012;U.S. Department of Education
Oﬃce for Civil Rights [OCR], 2014). Analyses of cross-sectional data from multiple years suggest that the risk of suspension of SWD
began increasing around 2000 (Krezmien, Leone, & Achilles, 2006;Zhang, Katsiyannis, & Herbst, 2004), possibly as SWD began
spending more of their school days in general education classrooms due to mainstreaming (U.S. Department of Education, 2016a,
2016b). General education teachers have been shown to be more likely to self-report attitudes of concern, indiﬀerence, or rejection
towards SWD than towards students without disabilities (Cook, Cameron, & Tankersley, 2007;Cook, Tankersley, Cook, & Landrum,
2000). Among SWD, those identiﬁed as having emotional disturbances (ED), speciﬁc learning disabilities (SLD), or speech or lan-
guage impairments (S/LI) have been reported to be more likely to be suspended (Achilles et al., 2007;Krezmien et al., 2006;Sullivan
et al., 2013). Racial and ethnic minorities with disabilities have been reported to be especially likely to be suspended (Achilles et al.,
2007;GAO, 2018;Krezmien et al., 2006;Losen & Gillespie, 2012), and so to experience the school-to-prison pipeline (Behnken et al.,
2014). This may occur because of a lack of training in eﬀectively managing the classroom behaviors of SWD, especially those who are
more likely to engage in behaviors that teachers view as disruptive or diﬃcult to manage. This may also occur because of a lack of
cultural or language sensitivity in working with SWD who are racial or ethnic minorities (Mendez & Knoﬀ, 2003;Zhang et al., 2004).
To limit the possibility that SWD who are racial or ethnic minorities are being suspended due to discriminatory discipline practices,
federal legislation requires U.S. states to monitor for signiﬁcant disproportionality in the use of suspension by schools districts. Recently
enacted federal regulations expand this monitoring by requiring school districts to use a standard methodology when monitoring for
signiﬁcant disproportionality in the use of suspension and other disciplinary practices (U.S. Department of Education Equity in IDEA
Rule, 2016b). These new federal regulations expand the manifestation determinations and other legal protections already provided
through the Individuals with Disabilities Education Act (IDEA) that ensure that schools continue to provide special education services to
SWD whose disruptive behaviors are related to underlying conditions (U.S. Department of Education Equity in IDEA Rule, 2016b). If
signiﬁcant disproportionality in the use of suspension is found using the standard methodology for SWD who are racial or ethnic
minorities, school districts in the U.S. are required to review and revise the policies, practices, and procedures that contribute to the
disparities as well as reallocate their federal funding (U.S. Department of Education Equity in IDEA Rule, 2016b). Federal im-
plementation of these regulations was recently delayed pending further review, including regarding whether the disparities are due to
discriminatory practices or instead to other explanatory factors (U.S. Department of Education, 2018a, 2018b).
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
1.2. Limitations in the knowledge base regarding whether and which SWD are being disproportionately suspended
Current empirical evidence regarding the extent to which SWD, including those who are racial or ethnic minorities, are being
disproportionately suspended as they attend U.S. schools is currently limited in several ways (Mendez & Knoﬀ, 2003). First, most
prior studies investigating whether SWD are disproportionately suspended have used limited statistical controls (e.g., year, race)
(Krezmien et al., 2006;Vincent, Sprague, & Tobin, 2012;Zhang et al., 2004) or cross-sectional designs (Anderson & Ritter, 2017;
GAO, 2018;Goran & Gage, 2011). Thus, the resulting risk estimates have been unadjusted for strong potential confounds (e.g.,
gender, age, family socioeconomic status [SES], prior behavioral or academic functioning). The risk attributed to having a disability
may therefore be partly—or fully—explained by other factors (Duran, Zhou, Frew, Kwok, & Benz, 2011;Theriot et al., 2010).
Second, very few studies have attempted to identify risk factors for more frequent suspension. Most of the available studies
instead have estimated the risk of being suspended dichotomously (i.e., “yes” or “no”) (e.g., Achilles et al., 2007;Bowman-Perrott
et al., 2013;Krezmien et al., 2006;Duran et al., 2011;Hemphill et al., 2009, 2012;Martin, 2014). Yet dichotomizing a variable that
could be measured continuously results in lost data, larger standard errors, reduced statistical power, and possibly biased risk
estimates (Altman & Royston, 2006;MacCallum, Zhang, Preacher, & Rucker, 2002). Identifying risk factors for more frequent sus-
pension is also substantively important. This is because students who are more frequently suspended are especially likely to ex-
perience (a) academic diﬃculties in school due to increased absenteeism and (b) involvement with the criminal justice system due to
less adult supervision and more opportunities to interact with criminally active individuals (Anderson & Ritter, 2017). Mowen and
Brent (2016) recently reported a strong relation between the frequency of suspension and children's risk of later arrest. Those who
were suspended across two, three, or four of the study's survey waves were 136%, 252%, and 400% more likely, respectively, to later
be arrested compared to those suspended only during one survey wave.
Third, and although a few available studies have used more extensive statistical controls, these studies have analyzed single
district- (Sullivan et al., 2013) or state-level (Anderson & Ritter, 2017;Kinsler, 2011) samples not designed to generalize to the U.S.
school-aged population. The few studies that have estimated the frequency of suspension among SWD ﬁnd that those with the speciﬁc
conditions of SLD, ED, or other health impairment (OHI), as well as students who are Black, male, or from lower SES families are
suspended more frequently. However, generalizability of the samples and statistical control for potential confounds have been
limited. For example, Sullivan et al.'s (2013) analyses indicating that SWD were more likely to experience multiple suspensions were
based on data from a single, unrepresentative school district in Wisconsin where 50% of the parents had at least a college degree.
Further, the cross-sectional data available to the investigators meant that prior behavioral functioning could not be controlled for as a
potential confound. Limited generalizability and lack of control for potential confounds also characterize the study by Sullivan, Van
Norman, and Klingbeil (2014) that analyzed data from the same district in Wisconsin. Both of these studies examined a constrained
range of 0, 1, or > 1 suspensions. Neither Mowen and Brent's (2016), nor Petras et al.'s (2011), nor Mendez and Knoﬀ's (2003)
analytical samples included SWD. Anderson and Ritter (2017) examined school-level but not individual-level risk factors for more
Because prior studies reporting on the suspension risk of SWD have been unable to adjust for prior behavioral functioning,
academic achievement, and other potential confounds (Sullivan et al., 2014), contrasts between otherwise similar students with and
without disabilities typically have not been possible. Yet contrasts between similarly situated students who diﬀer in their disability
status would better evaluate whether schools are discriminating against SWD when using suspension (e.g., NRC, 2004;OCR, 2016).
The very few studies reporting estimates designed to generalize to the U.S. school-aged population, which might be expected to best
inform federal legislation and policy, have largely examined whether suspension risk varies by speciﬁc disability condition but only
among samples of SWD (Achilles et al., 2007;Bowman-Perrott et al., 2013;Duran et al., 2011). Thus, it is currently unknown
whether, among students who are otherwise similar in their behavior, academic achievement, and additional background char-
acteristics, schools suspend those with disabilities more frequently. It is also unknown how suspension frequency varies by speciﬁc
disability conditions as well as how disability status and race or ethnicity interact.
Establishing whether and to what extent SWD, including those who are racial or ethnic minorities, attending U.S. elementary and
middle schools are being more frequently suspended than otherwise similar students without disabilities has important implications for
federal legislation and policy as well as for educational research and practice. Doing so would better clarify whether, as hypothesized
(Losen & Gillespie, 2012) and thought to be occurring widely (U.S. Department of Education Equity in IDEA Rule, 2016b), U.S. schools
are discriminating against SWD in their use of suspension. This would be particularly important to examine for those SWD who are
racial or ethnic minorities. If SWD who are racial or ethnic minorities are being more frequently suspended than SWD who are White,
this would suggest the need for greater use of culturally and language sensitive behavioral interventions by school psychologists, special
education teachers, and other practitioners to better meet the speciﬁc academic or behavioral needs of students with special needs. More
broadly, such an investigation would help establish whether, as has been reported to occur for the speciﬁc condition of attention-deﬁcit/
hyperactivity disorder (ADHD), being identiﬁed as disabled increases the likelihood of entering the metaphorical school-to-prison
pipeline for students who are racial or ethnic minorities through an attendant greater risk for suspension (Behnken et al., 2014). Finding
this to be the case would suggest that SWD who are racial or ethnic minorities may beneﬁt from additional academic and behavioral
supports to avoid suspension's associated life-course adversities. Investigating which students with speciﬁc disability conditions are
more frequently suspended would also better inform federally mandated monitoring eﬀorts.
1.3. Purpose of study
Given the aforementioned limitations of prior empirical work, the purpose of the current study was to evaluate the extent to
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
which SWD are being more frequently suspended as they attend U.S. elementary and middle schools. To address substantive and
methodological limitations in prior studies, we estimated the risk of being more frequently suspended using negative binomial
regression modeling of the number of times suspended by eighth grade from a nationally representative and longitudinal sample. We
statistically adjusted the risk factor estimates simultaneously for an extensive set of individual student-, family-, and school-level
covariates. Doing so should better approximate contrasts between similarly situated students and so yield rigorously derived esti-
mates of the risk for more frequent suspensions attributable to speciﬁc factors including disability status. We examined the following
1. Are students identiﬁed as having disabilities by the end of ﬁrst grade at greater risk of being more frequently suspended by the end
of eighth grade than students not so identiﬁed? Based on prior work (e.g., Sullivan et al., 2013), we hypothesized that SWD would
initially (i.e., in unadjusted estimates) receive more suspensions than students without disabilities.
2. To what extent is any initially observed risk of more frequent suspension for SWD explained by other student- (e.g., race/ethnicity,
gender), family- (SES), or school-level factors (racial, ethnic, and economic segregation), as well as plausibly exogenous individual
student-level behavior and achievement measured at school entry? Based on prior work investigating for potential discriminatory
disability identiﬁcation practices by schools (Hibel, Farkas, & Morgan, 2010;Morgan, Farkas, Hillemeier, & Maczuga, 2017), we
hypothesized that other explanatory factors would fully explain initial but unadjusted associations between having a disability
and the frequency of suspension. We expected that these other explanatory factors would include individual student-level race or
ethnicity, gender, family SES, prior history of externalizing problem behaviors, diﬃculties with attention and other learning-
related behaviors, academic achievement, and school-level racial, ethnic, and economic segregation. We also hypothesized that
these alternative explanatory factors would explain the greater suspension risk previously reported for students with speciﬁc
disability conditions including SLD, ED, and S/LI (Bowman-Perrott et al., 2013;Sullivan et al., 2014).
3. Are SWD who are racial or ethnic minorities suspended more frequently than SWD who are White? We hypothesized that other
explanatory factors (e.g., family SES, prior behavioral functioning) would also account for these interactions (Morgan et al.,
2.1. Data and analytical sample
We analyzed data from the Early Childhood Longitudinal Study-Kindergarten Class of 1998–1999 (ECLS-K), a data set collected
and administered by the National Center for Education Statistics (NCES). The ECLS-K is a nationally representative, longitudinal
cohort of children who were followed from kindergarten entry through the end of eighth grade. A multistage, probability sampling
design was employed in which approximately 1300 public and private schools were sampled from 100 geographic regions. About 24
kindergarten students were recruited from each school. Data collection continued throughout elementary and middle school. Surveys
were administered in kindergarten, ﬁrst, third, ﬁfth, and eighth grade. The NCES provided sampling weights to account for attrition
across sample waves. The study's sample contained information on 6740 students who participated in the ECLS-K from kindergarten
through eighth grade. As described below, we used multiple imputation to account for missing data for this sample.
2.2.1. School suspension
The dependent variable of interest was the number of times a student had been suspended from school by the spring of eighth
grade. During the parent interview portion of the ECLS-K eighth grade assessment, each parent was asked the number of times his or
her child had received an in- or out-of-school suspension. This was reported by parents as a cumulative total regardless of the ages or
grades at which each suspension occurred. Because this survey was administered to parents during the spring of the student's eighth
grade year, we refer to this variable as the student's number of suspensions through eighth grade.
2.2.2. Disability status
School personnel reported whether an Individualized Education Program (IEP) was on ﬁle. We considered SWD as those students
with IEPs on ﬁle who were receiving special education services due to formally identiﬁed disabilities. Along with the study's cov-
ariates, we used disability status (as indicated by an IEP being reported on ﬁle at the school) to predict the number of times students
were suspended by the spring of eighth grade. To reduce the possibility of reverse causality (i.e., that suspended students missed
instruction and thus performed less well academically, leading to being identiﬁed as disabled, such that any relation between these
variables was due to suspension resulting in disability identiﬁcation rather than vice versa), we conservatively measured disability
status as having been identiﬁed by the spring of ﬁrst grade. We did so because students identiﬁed as disabled received this desig-
nation prior to the preponderance of the time period when they were likely to be suspended (i.e., second to eighth grade). Doing so
helped ensure that any signiﬁcant relations between disability status and suspension frequency were most likely due to earlier
disability identiﬁcation aﬀecting later suspension frequency. Prior work has found that most suspensions occur during middle school,
followed by high school. Very few suspensions occur during elementary school, suggesting that even fewer occur by ﬁrst grade
(Mendez & Knoﬀ, 2003).
Students who had an IEP by the end of ﬁrst grade (i.e., 6.8% of the study's sample) represented only a subset of all students who
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
had an IEP by the spring of eighth grade. However, and by restricting attention to students identiﬁed as having disabilities by the end
of ﬁrst grade, we limited the possibility that any relations observed between disability status and suspension frequency were the
result of more frequently suspended students having been, as a consequence of suspension, identiﬁed as having disabilities. Further,
students with an IEP by ﬁrst grade should have had more severe impairments due to their earlier identiﬁcation and so displayed
greater academic or behavioral diﬃculties than students identiﬁed as having disabilities in the later grades. This made for a more
conservative test of whether disability status increased the frequency of suspension. This is because, if having a disability increased
the risk of suspension frequency, then student with more severe impairments should have been the most likely to be suspended more
frequently. Put another way, if suspension frequency was not elevated for SWD with more severe impairments, it seems unlikely that
suspension frequency would be elevated for SWD with less severe impairments. We also examined the risks of more frequent sus-
pension associated with speciﬁc disability conditions. These speciﬁc conditions included SLD, ED, S/LI, intellectual disability (ID),
OHI, developmental delay (DD), or other disability condition (ODC). We coded ODC as including the following rare disability
conditions: deaf-blindness, deafness, hearing impairment, orthopedic impairment, traumatic brain injury, and visual impairment.
2.2.3. Head Start
Participating in Head Start has been reported to increase the risk for suspension by middle school, although this risk may itself
result from children who participated in Head Start being more likely to later attend under-resourced schools (Aughinbaugh, 2001).
We, therefore, used participation in Head Start as a covariate when examining whether SWD were more likely to be frequently
suspended. Parents reported in the fall of kindergarten whether or not their child had been enrolled in federally sponsored Head Start.
We used a dummy variable for having been previously enrolled in Head Start (1 = yes, 0 = no).
2.2.4. Non-English-speaking household
Students with limited English proﬁciency have been reported to be at decreased risk of suspension (Anderson & Ritter, 2017).
Parents were asked about the primary language spoken in their home during the fall kindergarten survey. We coded those students
whose parents reported a primary language other than English as “1” and those students reported as using English as their primary
language as “0.” We used the kindergarten measure of this variable to help ensure that it was measured prior to the time period when
students were at risk for suspension.
2.2.5. Reading achievement
Greater academic achievement has been reported to decrease the risk for suspension (Arcia, 2006;Mizel et al., 2016). We ac-
counted for reading achievement as a covariate using the ECLS-K Reading Test administered during the fall of kindergarten. All items
were ﬁeld-tested and psychometric properties of items were evaluated using item response theory (IRT) methods. Items assessed in
kindergarten consisted of basic skills such as print familiarity, letter recognition, beginning and ending sounds, and rhyming sounds.
The reliability of the IRT-scale Reading Test scores for the kindergarten assessment was 0.93 (Pollack, Atkins-Burnett, Rock, & Weiss,
2005). We used the fall of kindergarten measure of this variable to ensure that it was measured prior to when students were most
likely to be suspended.
2.2.6. Mathematics achievement
Mathematics achievement in kindergarten was measured by the ECLS-K Mathematics Test. As with the Reading Test, all items
were ﬁeld-tested and their psychometric properties evaluated using IRT methods. The Mathematics Test's items during kindergarten
emphasized basic skills such as identifying numbers and shapes and counting objects. The reliability of the IRT-scaled Mathematics
Test scores in the kindergarten assessment was 0.92 (Pollack et al., 2005). As with the reading assessment, we used the fall of
kindergarten measure of this variable to ensure that it was measured prior to when students were most likely to be suspended.
Engaging in problem behaviors, particularly externalizing-type behaviors like ﬁghting or being disruptive, has been reported to
increase the risk for suspension (Mizel et al., 2016). Students are typically suspended due to recurrent acting-out behaviors (Skiba
et al., 2014). General education teachers completed the Social Rating Scale, a modiﬁed version of the Social Skills Rating System
(Gresham & Elliott, 1990) to rate an individual student's behavior in the fall of kindergarten. The teachers used a frequency scale to
rate how often the student displayed a particular social skill or behavior (i.e., 1 = never; 4 = very often). Items used for the Ap-
proaches to Learning subscale measured how well a student self-regulated his or her behavior while completing learning-related tasks
(e.g., attentive, task persistent, ﬂexible and organized) in the classroom. Controlling for prior achievement, these learning-related
behaviors best predict later achievement (Duncan et al., 2007;Tach & Farkas, 2006). The split-half reliability for the Approaches to
Learning subscale was 0.89 (Pollack et al., 2005). The Externalizing Problem Behaviors subscale measures the frequency of acting out
behaviors (e.g., argues with the teacher, ﬁghts, shows anger, disturbs the classroom). The split-half reliability for the Externalizing
Behaviors scale was 0.90 (Pollack et al., 2005). As with the achievement measures, we used the kindergarten measures of learning-
related and externalizing problem behaviors to ensure that they were measured prior to when students were likely to be suspended.
2.2.8. Socio-demographic characteristics
Students who are Black or Hispanic have been reported to have a higher risk of suspension (GAO, 2018;Petras et al., 2011), as
have boys (Sullivan et al., 2013). Students from lower SES families have also been reported to be more likely to be suspended than
students from higher SES families (Hemphill, Plenty, Herrenkohl, Toumbourou, & Catalano, 2014). Children's age, race or ethnicity
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
(coded as White, Black, Hispanic, or other), and gender have also been reported to increase the risk for suspension (Mizel et al.,
2016). Data on these variables were collected from parents during the ECLS-K kindergarten interviews. Family SES was assessed using
a multivariate parental self-report questionnaire measuring maternal, paternal, and/or guardian education levels and occupations as
well as family income at each of the survey waves. Because our dependent variable was the cumulative number of suspensions
received by the spring of eighth grade, we used the SES measure as experienced by students over this time period. Accordingly, we
followed the common practice of averaging these variables across the survey waves.
2.2.9. School characteristics
Attending a racially or economically segregated school has been reported to increase the risk of suspension (Hughes, Warren,
Stewart, Tomaskovic-Devey, & Mears, 2017). We therefore included covariates that captured information about the socio-
demographic composition of the schools that students attended including the percentages of students who were Black or Hispanic and
those eligible to receive free lunch. We averaged these variables across the survey waves so that each represented the combined
school-level racial and economic conditions experienced by the study's students.
We ﬁrst descriptively examined the data including calculating the percent with diﬀerent numbers of times suspended for students
with and without IEPs by ﬁrst grade. We then used negative binomial regression to model the number of times students were reported
to have been suspended. Allison (2012) recommends use of negative binomial regression models when examining count data. This is
because, unlike Poisson models that often ﬁt count data relatively poorly, negative binomial regression models allow for over-
dispersion (e.g., as here, where most students never experienced suspension and so there were a very large number of zero counts),
while also often being easier to estimate than zero-inﬂated Poisson models. Negative binomial regression also allows for greater
ﬂexibility when modeling variance (Winkelmann, 2003). Negative binomial regression models estimate the log of the expected count
as a function of the predictor variables. The resulting coeﬃcients are interpreted as follows: a 1-unit change in the independent
variable (e.g., having a disability) predicts a change in the diﬀerence in the logs of the expected counts of the dependent variable (i.e.,
being suspended) by the estimated coeﬃcient, holding the other predictor variables in the model constant.
We corrected the standard errors in these models for clustering of sample students in the kindergarten schools using the Huber-
White sandwich estimator, thereby adjusting for the non-independence of data from students in the same school when the sample was
drawn in kindergarten (Cameron & Miller, 2010;Primo, Jacobsmeier, & Milyo, 2007). We used SAS PROC SURVEYREG to obtain
these estimates. Multilevel models and Huber-White sandwich estimators provide similar values for standard errors as well as fully
adjust for clustering. Diﬀerence in use is explained more by disciplinary training (e.g., economics vs. psychology) than by speciﬁc
methodological advantages. For example, McNeish, Stapleton, and Silverman (2017) reported that “similar to HLM (hierarchical
linear modeling), the CR-SE (cluster-robust standard error) estimates fully address the clustered nature of the data” (p. 118).
Arceneaux and Nickerson (2009) reported that “clustered SEs, random eﬀects, and hierarchical models all adequately account for the
structure of the data” (p. 184). McNeish et al. (2017) further stated that cluster-adjusted standard errors made fewer assumptions
than multilevel models. Primo et al. (2007), when contrasting clustered standard errors to multilevel modeling, stated that “calcu-
lating clustered standard errors is a more straightforward and practical approach, especially when working with large datasets or
many cross-level interactions” (p. 446). Huang (2016) noted that use of cluster-adjusted standard errors based on Taylor linearization
is “well accepted, easy to implement, and is often considered the gold standard for variance estimation using complex sample data”
(p. 179). However, and as a robustness check, we re-estimated the negative binomial regression models using multi-level modeling
instead of CR-SE. The study's main results (available from the study's ﬁrst author) were consistent across both analytical methods.
We standardized the continuous predictor variables to facilitate comparison of eﬀect sizes. We used weights supplied by the NCES
to account for sample attrition by the eighth grade wave of the ECLS-K and multiple imputation to account for missing data. We used
Blimp software (Keller & Enders, 2017) when conducting the multiple imputation to account for the multilevel nature of the data
(Enders, Mistler, & Keller, 2016;Mistler & Enders, 2017). About 7% of the data were missing. We conservatively imputed 20 data sets
and then obtained parameter estimates for each. The resulting estimates were then averaged using the appropriate techniques in SAS
We addressed the ﬁrst research question by describing the detailed distribution of number of times suspended, separately for
students who did and did not have an IEP in ﬁrst grade. We then estimated two regression models. The ﬁrst simultaneous regression
had IEP by ﬁrst grade, the interaction between IEP and race or ethnicity, and a large number of control variables as additional
predictors. The second simultaneous regression repeated this calculation, but in place of whether the student had an IEP by ﬁrst grade
we used dummy variables for the student's speciﬁc disability condition. Use of simultaneous regression is recommended when, as is
the case here, the focus of the analyses is to assess the relative predicted eﬀects of a set of possible explanatory variables on a criterion
variable (Keith, 2015). We could not include interactions between speciﬁc disability conditions and race or ethnicity in this equation
because of small sample sizes for these combinations. Instead, we continued the inclusion of interactions between IEP or not and race
or ethnicity in the model. The results from these two models answer the study's second and third research questions.
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
3.1. Descriptive statistics
Table 1 shows the weighted descriptive statistics of the variables. In the sample of 6740 students, 16.6% had been suspended at
least once by the spring of eighth grade. Of those suspended, 10.1% had been suspended once, 3.2% were suspended twice, 1.7%
were suspended three times, 0.6% were suspended four times, and 0.9% were suspended ﬁve or more times. In the spring of ﬁrst
grade, 6.8% of the sample had an IEP. Of the entire sample, 0.7% percent had SLD, 0.2% had ED, 2.1% percent had S/LI, 0.3% had ID,
0.2% had OHI, 0.5% had DD, and 0.5% had ODC.
Table 2 shows the weighted distribution of the number of times students were suspended, separately by disability status in the
spring of ﬁrst grade. Overall, 20.1% and 16.3%, respectively, of students with and without disabilities in ﬁrst grade had been
suspended once or more by eighth grade. This answers the study's ﬁrst research question and is consistent with prior studies.
Speciﬁcally, when SWD are compared to students without disabilities and prior to controlling for other explanatory factors to make
the two groups otherwise similar, SWD descriptively had higher rates of suspension.
Descriptive statistics of selected variables (N= 6,740), weighted.
Percentage or M (SD)
Ever suspended by 8th grade 16.6%
Suspended 1 time by 8th grade 10.1%
Suspended 2 times by 8th grade 3.2%
Suspended 3 times by 8th grade 1.7%
Suspended 4 times by 8th grade 0.6%
Suspended 5 times by 8th grade 0.9%
IEP, spring 1st grade 6.8%
Other race/ethnicity 7.3%
Age, fall kindergarten 68.5 (4.4)
IEP, spring 1st grade, learning disabled 0.7%
IEP, spring 1st grade, emotional disturbance 0.2%
IEP, spring 1st grade, speech or language impairment 2.1%
IEP, spring 1st grade, intellectual disability 0.3%
IEP, spring 1st grade, other health impairment 0.2%
IEP, spring 1st grade, developmental delay 0.5%
IEP, spring 1st grade, other/missing 0.5%
Enrolled in Head Start, fall kindergarten 17.4%
From non-English household, fall kindergarten 11.9%
Family SES, average −0.0 (0.8)
Externalizing problem behaviors, fall kindergarten 1.6 (0.6)
Approaches to learning, fall kindergarten 3.0 (0.7)
Reading Test score, fall kindergarten, IRT-scale score 35.1 (10.1)
Mathematics Test score, fall kindergarten, IRT-scale score 26.4 (9.1)
School's percentage of Black students, average 16.0 (20.3)
School's percentage of Hispanic students, average 13.9 (18.7)
School's percentage of students receiving free lunch, average 33.1 (24.8)
Note. IEP = Individualized Education Program; IRT = item response theory; K = kindergarten;
SES = socioeconomic status. Sample size rounded as per NCES requirements.
Times suspended by the end of 8th grade, students with and without an IEP in ﬁrst grade and total sample, weighted.
Number of times suspended Students with IEP by spring of 1st grade
Students without an IEP by spring of 1st grade
0 79.9% 83.6% 83.4%
1 13.2% 9.9% 10.1%
2 3.1% 3.2% 3.2%
3 2.1% 1.6% 1.7%
4 0.9% 0.6% 0.6%
5 0.8% 1.0% 0.9%
Note. IEP = Individualized Education Program; sample sizes rounded as per NCES requirements.
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
3.2. Negative binomial regression model results
Table 3 shows two negative binomial regression models predicting the number of times students were suspended by the spring of
eighth grade. Model 1 shows no signiﬁcant coeﬃcient of IEP on suspension. The coeﬃcients of the interactions between IEP and race
or ethnicity were also non-signiﬁcant. Further, the coeﬃcients of the interactions between IEP and being Black or Hispanic were
negative. These results show that simultaneously controlling for an extensive set of alternative explanatory factors, we found no
support for the hypotheses that SWD were more frequently suspended than otherwise similar students without disabilities. We also
failed to ﬁnd that SWD who are Black or Hispanic or of other race/ethnicity were more frequently suspended than otherwise similar
SWD who are White.
In contrast, the coeﬃcient for students who are Black was positive and signiﬁcant. This suggests that students who are Black were
suspended more frequently than otherwise similar students who are White. However, this increased risk was unrelated to having a
disability as indicated by the non-signiﬁcant (and, for Black and Hispanic students, directionally inconsistent) interaction terms. The
negative binomial regression estimated the log of the number of events as an additive function of the predictors. Thus, to examine the
multiplicative eﬀect of a variable on the expected number of events, one should exponentiate its coeﬃcient. Exponentiating the
coeﬃcient for Black students in Model 1 of Table 3 gives exp. (0.45) = 1.57. Thus, and on average, students who are Black received
about 1.6 times as many suspensions by the end of eighth grade as otherwise similar students who are White. (We caution, however,
that suspension is a relatively rare event. As shown in Table 1, 83.4% of the sample were never suspended and an additional 10.1%
were suspended only once.)
As for the other variables in the equation, a number of them are statistically signiﬁcant. Males had a much higher rate of
suspension than females. Children who were older when they entered kindergarten had a higher rate than those who were younger.
Students from non-English-speaking households and from higher SES families were suspended less frequently. Kindergarten students
who engaged in externalizing problems more frequently were suspended more frequently during elementary and middle school.
Schools with a higher percentage of low-income students had a higher suspension rate.
Model 2 replaced the general IEP variable with the speciﬁc disability condition for which students were receiving special edu-
cation services. The IEP variable was deleted from this equation because it is redundant of having one of the speciﬁc disabilities.
(Accordingly, students within each disability category are being compared to students without disabilities.) In general, the estimates
Weighted parameter estimates of negative binomial regression models of the number of times suspended by the end of 8th grade
Model 1 Model 2
IEP in 1st grade −0.01 –
Hispanic −0.06 −0.05
Other race/ethnicity −0.33 −0.34
Age, fall kindergarten 0.11
Learning disabled 0.01
Emotional disturbance −0.14
Speech or language impairment 0.42
Intellectual disability 0.36
Other health impairment −1.46
Developmental delay −1.30
Other disability −2.31
IEP in 1st grade × Black −0.56 −0.53
IEP in 1st grade × Hispanic −0.22 −0.33
IEP in 1st grade × Other race/ethnicity 0.16 0.10
Enrolled in Head Start, fall kindergarten 0.13 0.12
From non-English household, fall kindergarten −0.65
Family SES, average −0.29
Externalizing problem behaviors, fall kindergarten 0.32
Approaches to learning, fall kindergarten −0.06 −0.06
Reading Test score, fall kindergarten −0.01 −0.01
Mathematics Test score, fall kindergarten −0.07 −0.07
School's percentage of Black students, average 0.06 0.05
School's percentage of Hispanic students, average 0.03 0.02
School's percentage of students receiving free lunch, average 0.21
Note. IEP =Individualized Education Program; SES = socioeconomic status. Multiple imputation and clustered standard errors
used. All models are clustered on fall kindergarten school. Twenty imputed BLIMP (Keller & Enders, 2017) datasets used.
Continuous variables standardized.
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
in this model are very similar to those in Model 1. As for the speciﬁc disability conditions, only ODC was statistically signiﬁcant. This
had a large and negative coeﬃcient. Thus, students with ODC were suspended less frequently during elementary and middle school
than students without disabilities, possibly because this variable includes conditions such as blindness and other physical disabilities
that may not result in disruptions of academic work in classrooms.
To summarize, the study's ﬁndings indicated that (a) even when disability status and many potential confounds were simulta-
neously controlled, students who are Black were more frequently suspended than otherwise similar students who are White; (b) the
interactions between race or ethnicity and disability status were not statistically signiﬁcant after controls including for family SES,
prior behavior, and school characteristics; (c) these interactions were directionally consistent with SWD who are Black or Hispanic
being less likely to be suspended than SWD who are White; and (d) the strongest relation between speciﬁc disability condition and
frequency of suspension occurred for students with ODC, who were less frequently suspended than students without disabilities.
3.3. Robustness check
As a robustness check, we also estimated three additional regression models (i.e., Poisson regression, zero-inﬂated negative
binomial regression, and zero-inﬂated Poisson regression, results available from the ﬁrst author). We found that Akaike's Information
Criteria (AIC) was lowest for the negative binomial regression of the four types of regressions that we estimated. Results from these
other models were consistent with results reported in Table 3's negative binomial regression. This suggested that the negative bi-
nomial regression results were robust to alternative regression model speciﬁcations.
Establishing whether and to what extent SWD are being suspended more frequently than otherwise similar students without
disabilities is an important issue of policy, research, and practice because of suspension's adverse life-course associations with lower
academic achievement, school dropout, substance abuse, juvenile delinquency, and adult criminality (e.g., Cuellar & Markowitz,
2015;Hemphill et al., 2012;Katsiyannis et al., 2013;Noltemeyer et al., 2015). For example, more frequent suspension greatly
increases the risk for juvenile arrest (Mowen & Brent, 2016), thereby serving as a potential entry point into the metaphorical school-
to-prison pipeline. Federal legislation and regulations require U.S. states to monitor for signiﬁcant disproportionality in suspension,
including for SWD who are racial or ethnic minorities (U.S. Department of Education Equity in IDEA Rule, 2016b). However, the ﬁeld
has lacked strong empirical evidence as to whether schools may be inappropriately suspending SWD more frequently than otherwise
similar students without disabilities. This is because few studies have estimated the risk of being suspended more frequently, in-
cluding as examined in analyses of a nationally representative sample that simultaneously accounted for a number of alternative
explanatory factors when estimating the risk attributable speciﬁcally to having a disability. Establishing that schools are suspending
SWD more frequently than otherwise similar students without disabilities would provide suggestive evidence that U.S. schools may be
using suspension in ways that are discriminatory (e.g., NRC, 2004; OCR, 2016).
In order to avoid potential reverse causality between SWD and suspension, we coded IEP as 1 only if the student had an IEP by
ﬁrst grade. Because very few suspensions occur prior to middle school (Mendez & Knoﬀ, 2003), these students were unlikely to have
been classiﬁed as having a disability because they had previously been suspended. By measuring the relation between this deﬁnition
of IEP and the number of times suspended through the spring of eighth grade, we were able to estimate the risk of having a disability
on suspension while limiting any eﬀect of suspension on the risk of having a disability.
Simple descriptive statistics showed that SWD have higher suspension rates than students without disabilities. However, multi-
variable regression models including controls for potential confounds fully explained this risk. Importantly, we found no signiﬁcant
interactions between IEP and race or ethnicity. Signiﬁcant risk factors for more frequent suspension included being Black, male,
older, raised in a English-speaking household, raised in a lower SES family, more frequently engaging in externalizing behavior
problems, and attending a school in which a higher percent of students in the school are from low-income families. For example, on
average, students who are Black receive about 1.6 times as many suspensions than otherwise similar students who are White across
elementary and middle school. However, and despite currently being an explicit target of compliance monitoring by federal legis-
lation and regulations (e.g., U.S. Department of Education Equity in IDEA Rule, 2016b), we found no evidence to suggest that SWD
who are racial or ethnic minorities are suspended more frequently than SWD who are White. Thus, and although students who are
Black are suspended more frequently, this risk is unrelated to whether they also are SWD.
Our study has several limitations. We report risk factor estimates. Our study does not allow for causal inferences. As in prior
studies examining for disproportionality in suspension (Wright, Morgan, Coyne, Beaver, & Barnes, 2014), including for SWD (e.g.,
Achilles et al., 2007;Bowman-Perrott et al., 2013;Duran et al., 2011), our analyses relied on a parent's retrospective report of school
suspension. Parents may not always have known how often their children had been suspended. However, and as seems likely, our risk
factor estimates are conservative if the resulting measurement error was random. It is also possible that we would have observed
other results if additional measures of the dependent variables (e.g., the number of days suspended for each disciplinary infraction,
whether the suspension was in- or out-of-school) and data sources (e.g., school records, student reports, or direct observations) had
Relatedly, data collection for the ECLS-K ended at the end of eighth grade. Although suspension occurs most frequently during
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
middle school (Mendez & Knoﬀ, 2003), we were unable to investigate whether the disparities that we observed continued to occur
during high school. We adjusted for prior behavior and achievement using measures administered during the fall of kindergarten,
which were plausibly exogenous to the occurrence of suspension throughout elementary and middle school. However, we may have
observed other results had we accounted diﬀerently for prior behavior. Alternatively, and because the ECLS-K's measure of sus-
pension surveyed parents about whether their children had been suspended by eighth grade, averaging the teacher behavioral ratings
across the elementary grades would have resulted in a more tenuous assumption of exogeneity. We did average the family's SES and
the racial and economic segregation of schools across kindergarten to eighth grade to account for the changing environmental
contexts that students experienced as they moved from the fall of kindergarten to the spring of eighth grade. The ECLS-K data were
collected until the spring of 2007. Analyses of more recently collected data are needed to replicate these ﬁndings.
4.2. Contributions and implications
Our ﬁndings have implications for federal legislation and policy as well as educational research and practice. We ﬁnd suggestive
evidence that U.S. elementary and middle schools use suspension in ways that may be discriminatory, at least as indicated by the
disparities in suspension frequency not being explained by measured confounds including student-level measures of behavior, family-
level SES, and school-level racial and economic composition. For example, and consistent with some studies (e.g., Petras et al., 2011;
Skiba et al., 2011), we ﬁnd that on average students who are Black are more frequently suspended than similarly situated students
who are White. Speciﬁcally, our analyses indicated that students who are Black receive about 1.6 times as many suspensions than
otherwise similar students who are White by the end of middle school. This increased risk for students who are Black for more
frequent suspension is not explained by potential confounds including student-level behavior, family SES, and school racial and
Variability in both sample characteristics and statistical analysis may explain diﬀerences between our ﬁndings and those of prior
work. For example, we analyzed a nationally representative sample in which family SES was directly assessed using parental surveys
of family income, occupation, and education level. In contrast, Kinsler (2011) analyzed state-level data and controlled for SES using
school records of receipt of free or reduced-price lunch. The ECLS-K's survey of parental education, occupation, and income, which
are averaged into a single composite variable, should have better controlled for family SES (Harwell & LeBeau, 2010), which is a
strong confound of race and ethnicity (Patten & Krogstad, 2015). Because the ECLS-K data included these more proximal measures of
family SES, our analyses should have better corrected for this strong confound and so allowed more rigorous estimates of the risk of
suspension attributable to race or ethnicity.
Our ﬁndings conﬂict with those reported in Wright et al.'s (2014) study, in which student suspension was dichotomously mea-
sured and most control variables including disability status were contemporaneously assessed at the eighth-grade survey while prior
behavioral functioning was averaged from kindergarten, ﬁrst, and third grade. In contrast, our study's individual-level predictor
variables were assessed in kindergarten and ﬁrst grade while the dependent variable was instead the number of times suspended by
the end of eighth grade. We identiﬁed students as having disabilities if they had an IEP by the spring of ﬁrst grade in order to limit the
possibility of reverse causality (i.e., suspension causing disability identiﬁcation). We therefore did not analyze the number of times
suspended for all students who ever had an IEP. Instead, we analyzed data from ﬁrst grade students who already had an IEP to
examine whether such identiﬁcation increased their likelihood of being suspended. We believe that this allowed for a conservative
analysis as this subgroup of students likely included an above average share of those with more severe impairments given their earlier
disability identiﬁcation. If SWD with more severe impairments were not at elevated risk for suspension, then it seems unlikely that
this would be the case for SWD with less severe impairments. We controlled for student-level behavior in kindergarten, which should
have temporally preceded suspension to further limit reverse causality. In addition, our family- and school-level variables were
averaged across all survey waves and so better accounted for changing environmental conditions students experienced from ﬁrst
through eighth grade. (Averaging across waves was also the best option to represent these environmental variables because the ECLS-
K parent survey did not ask about speciﬁc dates for suspensions.) One resulting practical implication of our study is that school
psychologists, teachers, and administrators should consider early interventions for kindergarten students already engaging in ex-
ternalizing problem behaviors. These students are at increased risk of being more frequently suspended throughout the elementary
and middle school grades.
Another practical implication of our study is that teachers, school psychologists, and administrators should consider whether
students are being suspended in ways that may discriminate based on sex, race, age, or economic background. This is because, as
others have found (Petras et al., 2011;Wright et al., 2014), socio-demographic disparities in suspension frequency are not fully
explained by variability in student-level externalizing problem behavior as well as other student-level indicators of school functioning
that might reasonably be related to the frequency of suspension. Educational practices that might help address these disparities
include increasing access to race- or gender-concordant teachers and/or those experienced in working with older students, as these
teachers may be able to advise on how to appropriately manage problem behavior in culturally sensitive ways that do not result in
suspension (Lindsay & Hart, 2017). Increasing the availability of evidence-based training of culturally sensitive practices may also be
helpful. School-to-home outreach eﬀorts to parents of children from traditionally marginalized populations may also help reduce
socio-demographic disparities in suspension frequency by helping stressed or under-resourced parents better manage their children's
externalizing problem behaviors as well as increasing the consistency of the behavior management approaches used across home and
school (Mason et al., 2016).
We ﬁnd little evidence that U.S. elementary and middle schools are suspending SWD more frequently than otherwise similar
students without disabilities. Our ﬁndings do not support policies or reports (Kim et al., 2010;Losen & Gillespie, 2012;OCR, 2014)
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
that having a disability is itself associated with an increased risk of entering the school-to-prison pipeline via school suspension
(Behnken et al., 2014;Mowen & Brent, 2016) conditional on this study's other explanatory factors. This includes SWD who are racial
or ethnic minorities, who were suspended no more frequently than SWD who are White. That SWD were not suspended more
frequently than students without disabilities as they attend U.S. elementary and middle schools conﬂicts with some prior work
(Krezmien et al., 2006), which did not adjust for potential confounds (Losen & Gillespie, 2012) including prior behavior (GAO, 2018;
Sullivan et al., 2013). However, our ﬁndings are consistent with other studies (e.g., Theriot et al., 2010), including those few that
have similarly accounted for the strong confound of prior behavior (Kinsler, 2011;Wright et al., 2014) and have also failed to ﬁnd
that SWD are more likely to be suspended than otherwise similar students without disabilities. We also fail to ﬁnd empirical evidence
to support federal legislation and policies (U.S. Department of Education Equity in IDEA Rule, 2016b) requiring monitoring for
signiﬁcant disproportionality in the extent to which SWD who are racial or ethnic minorities are being suspended as they attend U.S.
Achilles, G. M., McLaughlin, M. J., & Croninger, R. G. (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. https://doi.org/10.1177/10634266070150010401.
Allison, P. (2012 August 7). Do we really need zero-inﬂated models? Retrieved from https://statisticalhorizons.com/zero-inﬂated-models.
Altman, D. G., & Royston, P. (2006). The cost of dichotomizing continuous variables. BMJ, 332, 1080. https://doi.org/10.1136/bmj.332.7549.1080.
Anderson, K. P., & Ritter, G. W. (2017). Disparate use of exclusionary discipline: Evidence on inequities in school discipline from a U.S. state. Education Policy Analysis
Archives, 25, 1–32. https://doi.org/10.14507/epaa.25.2787.
Arceneaux, K., & Nickerson, D. W. (2009). Modeling uncertainty with clustered data: A comparison of methods. Political Analysis, 17, 177–190. https://doi.org/10.
Arcia, E. (2006). Achievement and enrollment status of suspended students. Education and Urban Society, 38, 359–369. https://doi.org/10.1177/0013124506286947.
Aughinbaugh, A. (2001). Does Head Start yield long-term beneﬁts? Journal of Human Resources, 36, 641–665. https://doi.org/10.2307/3069637.
Behnken, M. P., Abraham, W. T., Cutrona, C. E., Russell, D. W., Simons, R. L., & Gibbons, F. X. (2014). Linking early ADHD to adolescent and early adult outcomes
among African Americans. Journal of Criminal Justice, 42, 95–103. https://doi.org/10.1016/j.jcrimjus.2013.12.005.
Bowman-Perrott, L., Benz, M. R., Hsu, H.-Y., Kwok, O.-M., Eisterhold, L. A., & Zhang, D. (2013). Patterns and predictors of disciplinary exclusion over time: An analysis
of the SEELS national data set. Journal of Emotional and Behavioral Disorders, 21, 83–96. https://doi.org/10.1177/1063426611407501.
Cameron, C., & Miller, D. (2010). Robust inference with clustered data. In A. Ullah, & D. E. Giles (Eds.). Handbook of empirical economics and ﬁnance. CRC Press.
Carrell, S. E., & Hoekstra, M. L. (2010). Externalities in the classroom: How children exposed to domestic violence aﬀect everyone's kids. American Economic Journal:
Applied Economics, 2, 211–228. https://doi.org/10.1257/app.2.1.211.
Cook, B. G., Cameron, D. L., & Tankersley, M. (2007). Inclusive teachers' attitudinal ratings of their students with disabilities. The Journal of Special Education, 40,
Cook, B. G., Tankersley, M., Cook, L., & Landrum, T. J. (2000). Teachers' attitudes toward their included students with disabilities. Exceptional Children, 67, 115–135.
Cuellar, A. E., & Markowitz, S. (2015). School suspension and the school-to-prison pipeline. International Review of Law and Economics, 43, 98–106. https://doi.org/10.
Duncan, G. J., Dowsett, C. J., Claessens, A., Magnuson, K., Huston, A. C., Klebanov, P., ... Sexton, H. (2007). School readiness and later achievement. Developmental
Psychology, 43, 1428–1446. https://doi.org/10.1037/0012-1622.214.171.1248.
Duran, J. B., Zhou, Q., Frew, L.a., Kwok, O.-M., & Benz, M. R. (2011). Disciplinary exclusion and students with disabilities: The mediating role of social skills. Journal of
Disability Policy Studies, 24, 15–26. https://doi.org/10.1177/1044207311422908.
Enders, C. K., Mistler, S. A., & Keller, B. T. (2016). Multilevel multiple imputation: A review and evaluation of joint modeling and chained equations imputation.
Psychological Methods, 21(2), 222–240. https://doi.org/10.1037/met0000063.
Figlio, D. N. (2007). Boys named Sue: Disruptive children and their peers. Education Finance and Policy, 2, 376–394. https://doi.org/10.1162/edfp.2007.2.4.376.
Fletcher, J. (2010). Spillover eﬀects of inclusion of classmates with emotional problems on test scores in early elementary school. Journal of Policy Analysis and
Management, 29, 69–83. https://doi.org/10.1002/pam.20479.
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 and Treatment of Children, 34, 469–488. https://doi.org/10.1353/etc.2011.0035.
Gottfried, M. A., Egalite, A., & Kirksey, J. J. (2016). Does the presence of a classmate with emotional/behavioral disabilities link to other students' absences in
kindergarten? Early Childhood Research Quarterly, 36, 506–520. https://doi.org/10.1016/j.ecresq.2016.02.002.
Gregory, A., Skiba, R. J., & Noguera, P. A. (2010). The achievement gap and the discipline gap: Two sides of the same coin? Educational Researcher, 39, 59–68. https://
Gresham, P. M., & Elliott, S. N. (1990). Social Skills Rating System. Circle Pines, MN: American Guidance Service.
Harwell, M., & LeBeau, B. (2010). Student eligibility for a free lunch as an SES measure in education research. Educational Researcher, 39, 120–131. https://doi.org/10.
Hemphill, S.a., Heerde, J.a., Herrenkohl, T. I., Toumbourou, J. W., & Catalano, R. F. (2012). The impact of school suspension on student tobacco use: A longitudinal
study in Victoria, Australia, and Washington State, United States. Health Education & Behavior, 39, 45–56. https://doi.org/10.1177/1090198111406724.
Hemphill, S. A., Plenty, S. M., Herrenkohl, T. I., Toumbourou, J. W., & Catalano, R. F. (2014). Student and school factors associated with school suspension: A
multilevel analysis of students in Victoria, Australia and Washington State, United States. Children and Youth Services Review, 36, 187–194. https://doi.org/10.
Hemphill, S. A., Smith, R., Toumbourou, J. W., Herrenkohl, T. I., Catalano, R. F., McMorris, B. J., & Romaniuk, H. (2009). Modiﬁable determinants of youth violence in
Australia and the United States: A longitudinal study. Australian and New Zealand Journal of Criminology, 42, 289–309. https://doi.org/10.1375/acri.42.3.289.
Hibel, J., Farkas, G., & Morgan, P. L. (2010). Who is placed into special education? Sociology of Education, 83, 312–332. https://doi.org/10.1177/0038040710383518.
Horner, S. B., Fireman, G. D., & Wang, E. W. (2010). The relation of student behavior, peer status, race, and gender to decisions about school discipline using CHAID
decision trees and regression modeling. Journal of School Psychology, 48, 135–161. https://doi.org/10.1016/j.jsp.2009.12.001.
Horoi, I., & Ost, B. (2015). Disruptive peers and the estimation of teacher value added. Economics of Education Review, 49, 180–192. https://doi.org/10.1016/j.
Huang, F. L. (2016). Alternatives to multilevel modeling for the analysis of clustered data. The Journal of Experimental Education, 84, 175–196. https://doi.org/10.
Hughes, C., Warren, P. Y., Stewart, E. A., Tomaskovic-Devey, D., & Mears, D. P. (2017). Racial threat, intergroup contact, and school punishment. Journal of Research in
Crime and Delinquency.https://doi.org/10.1177/0022427816689811.
Katsiyannis, A., Thompson, M. P., Barrett, D. E., & Kingree, J. B. (2013). School predictors of violent criminality in adulthood: Findings from a nationally re-
presentative longitudinal study. Remedial and Special Education, 34, 205–214. https://doi.org/10.1177/0741932512448255.
Keith, T. Z. (2015). Multiple regression and beyond: An introduction to multiple regression and structural equation modeling (2nd ed.). New York: Routledge.
Keller, B. T., & Enders, C. K.. Blimp user's guide: Version 1.0. (2017). Downloaded from http://www.appliedmissingdata.com/blimpuserguide-4.pdf (on 2/15/2018).
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
Kim, C. Y., Losen, D. J., & Hewitt, D. T. (2010). The school-to-prison pipeline: Structuring legal reform. New York: NYU Press.
Kinsler, J. (2011). Understanding the black-white school discipline gap. Economics of Education Review, 30, 1370–1383. https://doi.org/10.1016/j.econedurev.2011.
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. https://doi.org/10.1177/10634266060140040501.
Lamont, J. H., Devore, C. D., Allison, M., Ancona, R., Barnett, S. E., Gunther, R., ... Guinn-Jones, M. (2013). Out-of-school suspension and expulsion. Pediatrics, 131,
Lindsay, C. A., & Hart, C. M. D. (2017, Winter). Teacher race and school discipline. Education Next, 17(1), 72–78. Retrieved from https://www.educationnext.org/ﬁles/
Losen, D. J., & Gillespie, J. (2012). Opportunities suspended: The disparate impact of disciplinary exclusion from school. Retrieved from https://civilrightsproject.ucla.
MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On the practice of dichotomization of quantitative variables. Psychological Methods, 7, 19–40.
Martin, A. J. (2014). The role of ADHD in academic adversity: Disentangling ADHD eﬀects from other personal and contextual factors. School Psychology Quarterly, 29,
Mason, W. A., January, S.-A. A., Fleming, C. B., Thompson, R. W., Parra, G. R., Haggerty, K. P., & Snyder, J. J. (2016). Parent training to reduce problem behaviors over
the transition to high school: Tests of indirect eﬀects through improved emotion regulation skills. Children and Youth Services Review, 61, 176–183. https://doi.org/
McNeish, D., Stapleton, L. M., & Silverman, R. D. (2017). On the unnecessary ubiquity of hierarchical linear modeling. Psychological Methods, 22, 114–140. https://doi.
Mendez, L., & Knoﬀ, H. (2003). Who gets suspended from school and why: A demographic analysis of schools and disciplinary infractions in a large school district.
Education and Treatment of Children, 26, 30–51.
Mistler, S. A., & Enders, C. K. (2017). A comparison of joint model and fully conditional speciﬁcation imputation for multilevel missing data. Journal of Educational and
Behavioral Statistics, 42(4), 432–466. https://doi.org/10.3102/1076998617690869.
Mittleman, J. (2018). A downward spiral? Childhood suspension and the path to juvenile arrest. Sociology of Education, 91, 183–204. https://doi.org/10.1177/
Mizel, M. L., Miles, J. N., Pedersen, E. R., Tucker, J. S., Ewing, B. A., & D'Amico, E. J. (2016). To educate or to incarcerate: Factors in disproportionality in school
discipline. Children and Youth Services Review, 70, 102–111. https://doi.org/10.1016/j.childyouth.2016.09.009.
Morgan, P. L., Farkas, G., Hillemeier, M. M., & Maczuga, S. (2017). Replicated evidence of racial and ethnic disparities in disability identiﬁcation in U.S. schools.
Educational Researcher, 46, 305–322. https://doi.org/10.3102/0013189X17726282.
Morris, E. W., & Perry, B. L. (2016). The punishment gap: School suspension and racial disparities in achievement. Social Problems, 63, 68–86. https://doi.org/10.
Mowen, T., & Brent, J. (2016). School discipline as a turning point: The cumulative eﬀect of suspension on arrest. Journal of Research in Crime and Delinquency, 53,
National Research Council (2004). Measuring racial discrimination. Panel on methods for assessing discrimination. Washington, DC: National Academies Press.
Neidell, M., & Waldfogel, J. (2010). Cognitive and noncognitive peer eﬀects in early education. The Review of Economics and Statistics, 92, 562–576.
Nicholson-Crotty, S., Birchmeier, Z., & Valentine, D. (2009). Exploring the impact of school discipline on racial disproportion in the juvenile justice system. Social
Science Quarterly, 90, 1003–1018. https://doi.org/10.1111/j.1540-6237.2009.00674.x.
Noltemeyer, A. L., Ward, R. M., & Mcloughlin, C. (2015). Relationship between school suspension and student outcomes: A meta-analysis. School Psychology Review, 44,
Patten, E., & Krogstad, J. M. (2015). Black child poverty rate holds steady, even as other groups see declines. Retrieved from http://www.pewresearch.org/fact-tank/
Petras, H., Masyn, K. E., Buckley, J.a., Ialongo, N. S., & Kellam, S. (2011). Who is most at risk for school removal? A multilevel discrete-time survival analysis of
individual- and context-level inﬂuences. Journal of Educational Psychology, 103, 223–237. https://doi.org/10.1037/a0021545.
Pollack, J., Atkins-Burnett, S., Rock, D., & Weiss, M. (2005). Early childhood longitudinal study, kindergarten class of 1998–99 (ECLS-K). Psychometric report for the
third grade (NCES 2005–062). U.S. Department of EducationWashington, DC: National Center for Education Statistics Retrieved from: https://nces.ed.gov/
Primo, D. M., Jacobsmeier, M., & Milyo, U. (2007). Estimating the impact of state policies and institutions with mixed-level data. State Politics and Policy Quarterly, 7,
Rumberger, R. W., & Losen, D. J. (2016). The high cost of harsh discipline and its disparate impact. Retrieved from https://www.civilrightsproject.ucla.edu/resources/
Skiba, R. J., Chung, C.-G., Trachok, M., Baker, T. L., Sheya, A., & Hughes, R. L. (2014). Parsing disciplinary disproportionality: Contributions of infraction, student, and
school characteristics to out-of-school suspension and expulsion. American Educational Research Journal, 51, 640–670. https://doi.org/10.3102/
Skiba, R. J., Horner, R. H., Chung, C.-G., Rausch, M. K., May, S. L., & Tobin, T. (2011). Race is not neutral: A national investigation of African American and Latino
disproportionality in school discipline. School Psychology Review, 40, 85–107.
Sullivan, A. L., Klingbeil, D. A., & Van Norman, E. R. (2013). Beyond behavior: Multilevel analysis of the inﬂuence of sociodemographics and school characteristics on
students' risk of suspension. School Psychology Review, 42, 99–114. https://doi.org/10.1177/1063426610377329.
Sullivan, A. L., Van Norman, E. R., & Klingbeil, D. A. (2014). Exclusionary discipline of students with disabilities: Student and school characteristics predicting
suspension. Remedial and Special Education, 35, 199–210. https://doi.org/10.1177/0741932513519825.
Tach, L. M., & Farkas, G. (2006). Learning-related behaviors, cognitive skills, and ability grouping when schooling begins. Social Science Research, 35, 1048–1079.
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. https://doi.org/10.1016/j.childyouth.2009.06.009.
U.S. Department of Education (2016a). School climate and discipline: Know the data. Retrieved from https://www2.ed.gov/policy/gen/guid/school-discipline/data.
U.S. Department of Education (2016b). Equity in IDEA Rule. Assistance to states for the education of children with disabilities; preschool grants for children with
disabilities. Retrieved from https://www2.ed.gov/policy/speced/reg/idea/part-b/idea-part-b-signiﬁcant-disproportionality-ﬁnal-regs-unoﬃcial-copy.pdf.
U.S. Department of Education (2018a). Assistance to states for education of children with disabilities: Preschool grants for children with disabilities, ﬁnal rule; delay of
compliance. Federal Register, 83(128), 31306–31317. Retrieved from https://www.regulations.gov/document?D=ED-2017-OSERS-0128-0393.
U.S. Department of Education (2018b). Regulation postponed two years to ensure eﬀective implementation. Retrieved from https://sites.ed.gov/idea/regulation-
U.S. Department of Education Oﬃce for Civil Rights (2014). Civil rights data collection data snapshot: School discipline. Retrieved from https://www2.ed.gov/about/
U.S. Department of Education Oﬃce of Civil Rights (2016). Dear colleague letter: Preventing racial discrimination in special education. Retrieved from https://www2.ed.
U.S. General Accountability Oﬃce. K-12 education: Discipline disparities for Black students, boys, and students with disabilities. (2018). Retrieved from https://www.gao.
gov/products/GAO-18-258 (Washington, DC (GAO-18-258)).
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13
Vincent, C. G., Sprague, J. R., & Tobin, T. J. (2012). Exclusionary discipline practices across students' racial/ethnic backgrounds and disability status: Findings from the
Paciﬁc Northwest. Education and Treatment of Children, 35, 585–601. https://doi.org/10.1353/etc.2012.0025.
Winkelmann, R. (2003). Econometric analysis of count data ((4th ed.)). Berlin, Germany: Springer.
Wright, J. P., Morgan, M. A., Coyne, M. A., Beaver, K. M., & Barnes, J. C. (2014). Prior problem behavior accounts for the racial gap in school suspensions. Journal of
Criminal Justice, 42, 257–266. https://doi.org/10.1016/j.jcrimjus.2014.01.001.
Zhang, D., Katsiyannis, A., & Herbst, M. (2004). Disciplinary exclusions in special education: A 4-year analysis. Behavioral Disorders, 29, 337–347. Retrieved from
P.L. Morgan et al. Journal of School Psychology 72 (2019) 1–13