Parsing the Relations of Race and Socioeconomic Status in Special Education Disproportionality

Abstract

This study investigated how student and school-level socioeconomic status (SES) measures predict students’ odds of being identified for special education, particularly high-incidence disabilities. Using the Early Childhood Longitudinal Study–Kindergarten cohort, hierarchical models were used to determine the relations of student and school SES to special education identification. Results indicated neither student-level SES variables for parent education, prestige, and income, nor school-level aggregates of SES measures, predicted overall special education placement, but higher parent education attainment was negatively related to high-incidence disability identification (adjusted odds ratio = 0.73). These findings suggest that racial disproportionality is not attributable to racial differences in income and indicate a need for further investigation of the mechanisms by which the longstanding racial disparities in special education emerge and are maintained. In particular, we discuss the implications of this study for further research into the relations of indicators of parent’s status on educational decisions within special education.

Keywords

socioeconomic status special education race disability disproportionality

Since Dunn’s (1968) commentary on the overrepresentation of students from marginalized cultural, linguistic, and economic backgrounds in special classes for students with disabilities, researchers have investigated both the extent of, and the factors that contribute to, racial disproportionality in special education. Disproportionality refers to extent group membership affects special education identification and outcomes (Coutinho, Oswald, & Best, 2002). These potential disparities are problematic because students who are mislabeled may receive inappropriate educational services, triggering questions about educational equity and the validity of educational decisions underpinning observed patterns. Two primary concerns drive resultant research and reform that some students—particularly those from culturally and linguistically diverse backgrounds—may receive unnecessary, ineffective, and stigmatizing special education services (Artiles, 1998), or, to a lesser extent, that students are denied needed services appropriate to their disabilities and unique educational needs (Bollmer, Bethel, Garrison-Mogren, & Brauen, 2007; Morgan et al., 2015). The effects of change efforts are questionable (see, for example, Albrecht, Skiba, Losen, Chung, & Middelberg, 2012), so ongoing policy and practice to reduce disparities may be bolstered by enhanced understanding of this complicated phenomenon, particularly the factors that drive disproportionality.

The nature and extent of disproportionality, and relations of race and socioeconomic status (SES) to special education identification in particular, remain highly controversial because of equivocal research findings and inconsistent application of child and school or community measures for SES (for examples and discussion, see Morgan et al., 2015; Skiba, Artiles, Kozleski, Losen, & Harry, 2016). Some researchers have argued that race is a proxy for SES and that SES may explain observed racial disparities (Hosp & Reschly, 2004), but this suggestion remains contentious. The complex relationships of these factors and race to special education identification have been insufficiently explained and often lack clear conceptualization or theoretical framing. That is, researchers often do not describe the conceptual or theoretical bases for their studies, selected measures, or interpretations (Skiba et al., 2016; Sullivan & Artiles, 2011), which contributes to inconsistency in the operationalization of SES and interpretation of findings (Harwell & LeBeau, 2010). In the present study, we attempt to help clarify this debate by adopting an explicit conceptualization of SES to examine how both student- and school-level SES variables contribute to students’ special education identification. Below, we synthesize two relevant literatures: findings on relations of SES to special education disproportionality and sociological conceptualizations of SES, followed by description of the purpose of the present study in exploring whether measures based on a tripartite model of SES extend our understanding of disproportionality in special education.

Relations of SES to Disproportionality

Research on patterns of disproportionality is increasingly fraught with contradictions, particularly across important sampling dimensions such as unit of analysis (i.e., student, school, district), geographic context (e.g., district, state, region, nation), and the specific covariates investigated. Divergent findings have sparked intense debate given the conflicting, and potentially deeply problematic, implications for policy and practice (for examples and recent discussions, see Cohen, Burns, Riley-Tillman, & Hosp, 2015; Morgan et al., 2015; Skiba et al., 2016). The relations of SES and race in this debate are often singled out as some scholars attribute observed racial disparities to the potential environmental and neurobiological sequelae of poverty; however, findings across the body of disproportionality studies seem to preclude such simplistic explanations (see Skiba et al., 2016; Skiba, Poloni-Staudinger, Simmons, Feggins-Azziz, & Chung, 2005, for discussion).

The variability in operationalization of SES in this literature complicates interpretation and synthesis of findings. A variety of measures have been used, and rationales for the selection of variables are rarely provided. Variables selected range from students’ free/reduced lunch (FRL) status (Sullivan & Bal, 2013) or family’s composite SES (Hibel, Farkas, & Morgan, 2010; Hibel & Jasper, 2012; Morgan et al., 2015), marital status, and maternal age (e.g., Blair & Scott, 2002), to school or community aggregates for FRL (Sullivan & Artiles, 2011), rates of high school completion, median household income, median property values, and percentage of households below poverty (Coutinho et al., 2002; Eitle, 2002; Hosp & Reschly, 2004; Oswald, Coutinho, & Best, 2002; Oswald, Coutinho, Best, & Nguyen, 2001). Historically, most studies have utilized school, district, and state aggregates to study patterns and predictors of disproportionality, and the relations of SES measures to disproportionality have been weak and inconsistent. In some cases, these measures for SES have been found to be positively related to overidentification (e.g., learning disability [LD] in Blair & Scott, 2002; Sullivan & Artiles, 2011; intellectual disability [ID] in Eitle, 2002; LD and ID in Oswald et al., 2001; LD, ID, and emotional disability [ED] in Hosp & Reschly, 2004), whereas others have found an inverse relationship to identification (e.g., ID in Coutinho et al., 2002; Oswald et al., 2002; Oswald et al., 2001; ED in Oswald, Coutinho, Best, & Singh, 1999; Skiba et al., 2005), and still others no relations (ED and ID in Sullivan & Artiles, 2011), with patterns varying depending on the racial group considered.

A common limitation of many early studies of disproportionality is that researchers often relied on district-level aggregates for outcomes and covariates, rather than student-level data, which may not accurately capture the effects of focal variables (Sullivan & Bal, 2013). More recent studies have utilized student data from state and nationally representative samples (e.g., Shifrer, Muller, & Callahan, 2010) and applied multilevel models (Hibel et al., 2010; Sullivan & Bal, 2013), but as with earlier research, findings have been conflicting. Shifrer and colleagues (2010) found that parent income, but not parent education, was related to students’ risk of identification as LD among secondary students. Hibel and colleagues (2010) found that neither family nor school composite SES, measured as a standardized scale, was a predictor of special education identification at fifth grade when other student and school-level covariates were included. Sullivan and Bal (2013) found that parents’ education was inversely related to special education risk and students’ FRL status was positively related to risk, but that the proportion of students in the school who received FRL, a common variable used in earlier studies, was not. Most recently, Morgan and colleagues (2015) included a composite SES measure in their study of disproportionality and found no relation to students’ risk of identification for LD, ID, ED, or speech-language impairment (SLI), although it was related to identification for other health impairments (OHI).

Although this recent research advances earlier studies that failed to account for students’ SES, the variability in measures of SES and the lack of explicit theorization in the operationalization of SES continues to hinder interpretation of what observed relations might mean. Explicitly conceptualized variables allow researchers to evaluate the applicability of theories to disproportionality and potential intervention and reform efforts. For instance, some scholars suggest that special education disproportionality is partially attributable to minority children’s increased risk of exposure to hazardous environmental variables in impoverished settings (e.g., malnutrition, lead paint in older dwellings; Donovan & Cross, 2002; Hosp & Reschly, 2004; MacMillan & Reschly, 1998). Other scholars believe that disproportionality emerges from cultural and class differences between a predominantly White middle class teaching force and a more diverse student body (e.g., Artiles, 1998). Neither explanation has been consistently supported by the research but they are also rarely studied directly; instead, variables such as FRL status or SES composites are used. This general limitation is likely related to a common methodological issue in the earlier literature: reliance on secondary data and researchers’ tendency to use available variables to operationalize SES rather than offering a conceptual model for study and selecting or constructing measures accordingly (Bollen, Glanville, & Stecklov, 2001; Harwell & LeBeau, 2010).

These common measures have several weaknesses. The FRL status is among the most popular because it is typically inexpensive to obtain. However, FRL is a blunt, flawed measure of SES confounded by several problems including unrelated administrative issues that reduce accuracy (Harwell & LeBeau, 2010) and decreased participation rates as students get older which may be due to missing paperwork, increased income, or stigma of using FRL. Using community-level SES measures alone are problematic because they assume the student population in a given community is homogeneous, which may not reflect the variable’s relations to individual outcomes and invite the ecological fallacy in which student-level inferences are erroneously drawn from community-level data (Sullivan & Bal, 2013). Composite measures in particular can be problematic because they aggregate aspects of SES that may contribute differentially to outcomes (American Psychological Association [APA], 2007) and may result in smaller effect sizes than the individual measures (Bollen et al., 2001). In addition, the endogeneity of SES measures presents an added methodological problem when researchers use a single measure as a proxy for this construct. Disaggregating these variables results in more accurate estimation of SES by its individual components (APA, 2007). Taken together, student- and community-level data may capture both proximal and distal SES effects and enhance our theorization of mechanisms behind disproportionality. Aligning measures with one or more conceptual models or theories of SES would further strengthen estimations of the effects of SES on disproportionality.

The Tripartite Conceptualizations of SES

Despite its importance in the social sciences, particularly in how we understand disparities, there is no agreed upon definition of SES, although SES is broadly conceptualized as an indicator of resource access (J. M. Oakes & Rossi, 2003). This variability complicates its measurement and comparison across studies broadly, and within the disproportionality literature more specifically, but revisiting the theoretical roots of the study of SES and contemporary models for operationalizing SES may improve precision in measurement and interpretation. The concept of SES, and resultant theories of its relations to individuals’ outcomes, has its roots in sociology. The materialist approach draws on early conceptualizations of class focusing on individuals’ access to scarce resources, whereby inequities and social stratification emerge through efforts of the wealthy to maintain property and other wealth. From a materialistic perspective, social stratification is driven by property and wealth (Marx & Engels, 1848/2004). Thus, a materialistic analysis emphasizes the relations of income or other wealth to individuals’ outcomes (APA, 2007). From this perspective, minority students, coming from households with less wealth, on average, would be pushed in or kept out of special education by the wealthier mainstream groups depending on the perceived value of the services.

Over time, sociologists’ conceptualizations of class shifted to emphasize differences in not only wealth but also status that resulted in varied resources, behaviors, and beliefs across social strata. This gradient approach emphasizes relative differences between groups or classes (APA, 2007; Weber, 1920/2004), including resultant lifestyle differences and status or power differentials that produce and maintain inequality (Ridgeway, 2014). High-status individuals maintain their stratum position by monopolizing the resources and opportunities (Weber, 1920/2004). Thus, research grounded in this perspective often focuses on the influence of both status dimensions of class (e.g., education level and occupational status based on prestige or social standing) and the economic dimension of class (e.g., income).

A tripartite model of SES emphasizing the roles of income, education, and occupation or prestige gained traction in the 1970s (for discussion, see Sirin, 2005). Occupational status or prestige more closely aligns with Weber’s conceptualization of status, but educational attainment and income are often treated as more objective measures of SES (J. M. Oakes & Rossi, 2003). Perhaps as a result, in educational research, the oft available FRL is a common proxy for SES because eligibility for this federal program is based on family income. This measure, however, is fraught with limitations, particularly related to reliability, that undermine its utility in research (for discussion, see Harwell & LeBeau, 2010) in addition to underrepresenting SES. Conversely, another common approach is to rely on SES composites that combine the three dimensions into a single. Both FRL and composites are too coarse to allow for nuanced theorization of the mechanisms by which SES influences outcomes.

Consider, for instance, that an individual’s educational attainment, occupational prestige, and income may be incongruent (e.g., the adjunct professor with a PhD whose income places them below the federal poverty level), or that one dimension may be related to different mechanisms for change. Not surprisingly then, educational research grounded in this tripartite model of SES with corresponding measures of each dimension has shown that each uniquely contributes to outcomes and should be measured separately (Sirin, 2005). Likewise, in health research, various dimensions of SES are also differentially related to outcomes (e.g., Veensra, 2000) perhaps because they confer different resources and opportunities (e.g., income facilitating access to care whereas education results in more healthful behaviors). Thus, there is value in considering the relations of each dimension of SES to disproportionality, where the discourse often engages both education and health disparities, rather than relying on composite measures or proxies of single dimensions.

The Present Study

Given the relevance of SES measures aligned with the tripartite model to differential education and health outcomes, we explored the relations of parental education, occupational prestige, and income to special education identification. More specifically, the purpose of this study was to investigate the effects of student- and school-level SES measures of education, prestige, and income on students’ identification for special education and in high-incidence disability categories, addressing the two research questions.

Research Question 1: To what extent do student-level SES measures moderate the relationship between race and overall special education identification and identification in the high-incidence categories?

Research Question 2: To what extent do school-level SES measures moderate the relationship between race and overall special education identification and identification in the high-incidence categories?

We examined both overall special education status and the high-incidence disabilities since the latter are often the focus of disproportionality research and reform, as together, these categories are considered to be those most susceptible to professional judgment and bias compared with the more medically or physically based low-incidence disabilities (Klingner et al., 2005). Both student and school-level covariates were included to model how the surrounding environment influences students’ SES.

Method

Data Source and Procedures

This study drew data from the Early Childhood Longitudinal Study–Kindergarten Cohort (ECLS-K), a nationally representative longitudinal study of approximately 21,000 students who entered kindergarten in 1998 in 1,591 U.S. schools. The ECLS-K was developed and sponsored by the U.S. Department of Education to study student achievement, developmental status, school quality, and the relationships of various contexts (e.g., family, early childhood education, school) to development and achievement (Tourangeau, Nord, Le, & Sorongon, 2009).

Sampling and weighting

As described in extensive detail by Tourangeau and colleagues (2009), the ECLS-K features,

multistage probability sample design to select a nationally representative sample of children attending kindergarten in 1998–99. In the base year the primary sampling units (PSUs) were geographic areas consisting of counties or groups of counties. The second stage units were schools within sampled PSUs. The third- and final-stage units were children within schools. (p. 4-1)

The population of 5-year-olds determined primary sampling unit (PSU) size, with oversampling of children identified as Asian or Pacific Islander to provide sufficient subgroup sample size. The study sampled from the largest 24 PSUs, and then the remaining 38 PSUs were stratified into equal size units based on metropolitan statistical area status, minority composition, population, and per capita income. To ensure representativeness, the designers “freshened” the sample for the first-grade data collection to include children who did not attend kindergarten, and followed subsamples of students who transferred schools, retaining all language minority students (pp. 4-1–4-2). The public and private schools sampled in the second stage included schools from the Bureau of Indian Affairs and Department of Defense domestic schools. Probability of school sampling was proportional to enrollment within the public and private sampling strata of all schools with kindergarten programs. In addition, designers applied a variety of “field and sampling procedures” to account for sample attrition expected in longitudinal studies (Tourangeau et al., 2009, p. 4-24).

Sampling weights and corresponding replicate weights (to estimate standard errors using the jackknife replication method) provided within ECLS-K correct for design effects, attrition, and patterns of nonresponse to allow for population estimates (Tourangeau, Nord, Le, & Wan, 2002). The sample weights account for differential selection probabilities across the four sampling stages, as well as nonresponse at the child, teacher, and school levels so that estimates using the weights are representative of all students, teachers, and kindergarten programs in the base year. Multiple sample weights and combination weights were computed for the various instruments administered. The designers recommended use of the combination weights from data are drawn from multiple instruments (Tourangeau et al., 2009). Given the measures used in the present study as described below, we used the C5CPTW0 sample weight to provide nationally representative estimates and robust standard errors.

Procedures

Data collection occurred periodically following a longitudinal design at fall and spring of kindergarten, a subsample the fall of first grade, and again for all participants during the springs of first, third, fifth, and eighth grades. Parents provided consent using paper consent forms, verbal consent via the phone, or home visits by ECLS researchers. Each wave of data collection utilized multiple instruments, including parent interviews, student assessments, questionnaires completed by various school personnel (i.e., teachers, special education staff, and school administrators), and review of school records.

Parent and special education teacher questionnaires were used in this study. Parent questionnaires were administered by trained staff via computer-assisted telephone and in-person interviews during each data collection wave. During these interviews, respondents provided information on family characteristics including race, parent’s highest level of education, family income, and professional status. Questionnaires were mailed to schools for teachers and school administrators to complete. Students’ disability categories were determined by special education teacher report. School records indicated receipt of special education (i.e., whether the student had an individualized education plans [IEPs] on record), and primary disability, as well as the student’s gender, race, and school location.

Analytic Sample

The analytic sample consisted of students participating in the third-grade data collection for whom demographic data were available (n = 8,500). Analyses excluded students who did not have data during the third wave data collection and for whom demographic and special education data were not available. Given that hazard ratios of special education identification peak in third grade (Hibel & Jasper, 2012; Morgan, Staff, Hillemeier, Farkas, & Maczuga, 2013) and that special education prevalence in the ECLS-K third-grade wave was consistent with national estimates of incidence of special education identification (National Center on Education Statistics [NCES], 2012), third grade emerged as an appropriate time to collect disability information. Descriptive characteristics for the unweighted sample and weighted population are provided in Table 1. Because of the hierarchical nature of the data (i.e., students nested within schools) both student data (Level 1 variables) and school data (Level 2 variables) were used. Preliminary analyses of differences between responders and nonresponders in a given year showed that the analytic sample was significantly more affluent and less racially diverse than the full unweighted sample of ECLS-K participants. However, we incorporated the C5CPTW0 sample weight to correct for study design effects, sampling, and variation in participation across the waves of data collection, thus providing nationally representative estimates in the present analysis. As such, we report results for the population (i.e., weighted sample).

Table 1.

Sample and Population Characteristics.

	Unweighted (n = 8,500)		Weighted (N = 3,001,250)
	n	%	n	%
Student characteristics
Gender
Boys	4,300	50.47	1,547,650	51.57
Girls	4,200	49.53	1,453,600	48.43
Special education
Yes	800	9.73	366,300	12.21
No	7,700	90.27	2,634,950	87.79
High incidence	300	3.58	141,550	4.72
Race
White	5,950	70.20	1,925,600	64.16
Black	800	9.32	421,750	14.05
Hispanic	1,200	14.24	554,450	18.47
Asian/PI	550	6.24	99,431	3.31
Parent characteristics
Education level
Less than high school	450	5.40	228,150	7.60
HS diploma	1,950	23.12	828,650	27.61
Some college	2,800	33.12	979,550	32.64
Bachelor’s	2,100	24.54	632,800	21.08
Advanced degree	1,200	13.82	332,100	11.07
Income
<25,000	1,650	19.43	782,150	26.06
25,000–50,000	3,000	35.03	1,072,500	35.74
>50,000	3,850	45.54	1,146,600	38.2

Level 1 Variables

Table 2 provides a tabular list of variables and their coding. Student’s sex was collected from school record reviews and dichotomized (0 = boy; 1 = girl). Student race was taken from parent interviews during spring of first grade. Four racial categories (White, Black, Hispanic, and Asian) were dummy-coded with White as the referent group. Students identified as American Indians could not be included in the present study because of low participation rates preclude calculation of reliable estimates, based on recommendations by the ECLS-K developers (Tourangeau et al., 2002). Determination of SES encompassed three theoretically derived components: the highest level of education obtained by a student’s parents (e.g., if the mother had a doctorate and the father had a high school diploma, the mother’s education level was used), family income, and professional status. Family income was represented by the ECLS-K imputed income for each family; despite a skewed income distribution, it did not affect the residuals in preliminary logistic regression models run in SPSS. Professional status was derived from prestige scores adapted from the General Social Survey (GSS; Nakao & Treas, 1990), which were computed by having a normative sample rate various occupations on a scale from 0 to 100, with higher ratings indicating greater prestige.

Table 2.

Variables Used in Analyses.

Variables	Description	Coding
Independent variables (Level 1)
Sex	Student’s sex as reported by school records	0 = male, 1 = female
Race	Parent reported student’s racial category as captured by ECLS. American Indians were excluded due to low participation rates	Three dummy-coded categories (Black, Hispanic, Asian) with White as the referent
Education	Parents’ highest level of education. For this variable, if the mother had a doctorate and the father had a bachelor’s degree, the parent’s highest education was recorded as an advanced degree (based on the mother’s doctorate).	0 = Less than a high school diploma 1 = High school diploma or equivalent 2 = Some college 3 = Bachelor’s degree 4 = Advanced degree
Income	A continuous variable of parent’s income for each family. Missing incomes were imputed by ECLS.	Continuous variable
Professional status	Prestige scores adapted from the General Social Survey (Nakao & Treas, 1990) computed by having a normative sample rate various occupations on a scale from 0 to 100, with higher ratings indicating greater prestige for a given position.	Continuous variable
Independent variables (Level 2)
Education	The average education level for a school as reported by parents during parent interviews	Mean of within-school reported parents’ highest education level
Income	The average income level for a school as reported by parents during parent interviews	Mean of within-school parental income based on Level 1 information
Professional status	The average prestige of parent’s professional positions based on the General Social Survey (Nakao & Treas, 1990) ratings	Mean of within-school professional status based on Level 1 information
Dependent variables
Special education status	Student’s special education status was determined by the presence of an IEP in a student’s third-grade school records.	0 = No special education receipt 1 = Special education receipt
High-incidence disability	Student’s special education teachers reported disability categories. If a student was identified with a learning disability, intellectual disability, or emotional-behavioral disability, they were included in the high-incidence disability category, with youth who had not been identified for services as the referent.	0 = No special education receipt 1 = LD, ID, or ED status

Note. ECLS = Early Childhood Longitudinal Study; IEP = individualized education plan; LD = learning disability; ID = intellectual disability; ED = emotional disability.

Special education status, as determined by the presence of an IEP, was taken from students’ third-grade school records and served as a dependent variable in the data analyses (1 = special education, 0 = not in special education). Students identified with LD, ID, or ED in their special education teacher questionnaires were coded as having a high-incidence disability (1 = yes, 0 = no), which also served as a dependent variable. The high-incidence categories—LD, ID, and ED—have received considerable attention in past disproportionality research because of the impact of professional judgment or bias (e.g., Gresham, Sugai, & Horner, 2001), with some scholars suggesting there are few meaningful differences in the students served in these categories (e.g., Sabornie, Cullinan, Osborne, & Brock, 2005). They have been combined here into a single dependent variable because the small frequencies of each disability by race did not allow for reliable estimates for individual categories. This decision to combine categories was based on recommendation from the ECLS-K developers on the minimum unweighted cell sizes required for reliable analyses (e.g., Tourangeau et al., 2002).

Level 2 Variables

The Level 2 SES variables mirror the Level 1 measures of SES and included the average parental occupational prestige, parent income, and education level of parents for that school.

Data Analysis

To verify the appropriateness of hierarchical modeling and inclusion of Level 2 factors, an unconditional model and two analytic models were fit. The unconditional model was η_ij = γ₀ + u_0j where $η_{ij}$ is the log odds (logit) of a student being placed in special education, γ₀ represents the average log odds that a student would be placed into special education, u_0j represents the unique school effect on a students’ log odds of being placed in special education. After establishing that there was variation between schools with respect to special education placement, the first analytic model included student sex and race at Level 1, continuing to allow school placement to vary using the following equation:

η_{ij} = γ_{00} + γ_{10} (Sex) + γ_{20} (Black) + γ_{30} (Hispanic) + γ_{40} (Asian) {+ u}_{0 j} .

The second analytic model added the three SES variables, group centered at the student level, with the averages of the Level 1 SES variables at the school level, represented by the following equation:

\begin{matrix} η_{ij} = γ_{00} + γ_{01} (Avg Income) + γ_{02} (Avg Prestige) + γ_{03} (Avg Education) + γ_{10} (Sex) \\ + γ_{20} (Black) + γ_{30} (Hispanic) + γ_{40} (Asian) + γ_{50} (ParEd) + γ_{60} (ParPrestige) + γ_{70} (ParIncome) {+ u}_{0 j} . \end{matrix}

To investigate disproportionality in high-incidence disabilities, three hierarchical models identical to Analysis 1, save for the dependent variable, were fit. Students not identified for special education served as the referent group for these models. The unconditional model was fit with the equation η_ij = γ₀ + u_0j where γ₀ represents the average log odds that a student will be placed in a high-incidence disability category, and u_0j represents the unique school effect on the student’s log odds of being placed in a high-incidence disability. Similarly, two analytic models that were identical to those in Analysis 1, except that the outcome of interest was the log odds of student identification in a high-incidence disability category, were fit.

Power Analysis

The ECLS-K analytic sample included 8,500 students in 1,521 schools. An a priori power analysis indicates whether the data provide adequate power to detect a relationship between minority status and special education risk after taking SES and sex into account. The Optimal Design software (Raudenbush, 2011) was used to determine the power of the unweighted analytic sample. Because the analytic sample size (i.e., number of students and schools) is already known, the power analysis aimed to identify the level of power the sample would yield with the probability of being placed in special education (0.12), an alpha level of .05, with six students in each school. The power analysis indicated that due to the large number of schools, power exceeded .99 regardless of how small the number of students in each school became. The power still exceeded .90 if the probability of being placed in special education dropped to half of that suggested by the literature (.06).

Results

Special Education Identification

To establish whether a hierarchical model was warranted, a fully unconditional model was fit to the data to see whether special education identification varied between schools. There was a significant amount of variation between schools in special education identification (χ² = 2847.42, df = 1521, p < .001; intraclass correlation coefficient [ICC] = 0.25; Snijders & Bosker, 1999). Consequently, additional models were fit to assess the effects of minority and socioeconomic factors on disproportionality in special education across schools. As shown in Table 3, the first analytic model included student’s race and sex as covariates. Girls were less likely to be placed in special education (adjusted odds ratio [AOR] = 0.51), or, equivalently ${(. 51)}^{- 1} = 1.96$ meaning boys were almost twice as likely to be placed in special education. Asian Americans were significantly less likely to be placed in special education than White students (AOR = 0.30). The second analytic model included student sex and race, and added parent education, prestige, and income at the student level. At the school level, the average education level, prestige score, and income were included. The additional covariates did not have a significant effect on student identification in special education. Both girls (AOR = 0.52) and Asian Americans (AOR = 0.30) continued to be significantly underrepresented in special education identification. Slope variation was constrained to zero since preliminary analyses indicated that there was no significant variation in slopes between schools.

Table 3.

Student Odds of Being Placed in Special Education.

	Model 1			Model 2
	β	AOR	CI	β	AOR	CI
Intercept	−1.84	0.16	[0.13, 0.19]	−1.70	0.18	[0.06, 0.56]
Student level
Girls	−0.67*	0.51	[0.40, 0.65]	−0.66*	0.52	[0.41, 0.65]
Black	0.07	1.08	[0.70, 1.66]	−0.14	0.87	[0.58, 1.32]
Hispanic	0.12	1.12	[0.80, 1.58]	−0.17	0.85	[0.59, 1.22]
Asian	−1.19*	0.30	[0.17, 0.56]	−1.21*	0.30	[0.14, 0.63]
Education				−0.19	0.82	[0.72, 0.95]
Prestige				−0.01	1.00	[0.99, 1.01]
Income				−0.01	1.00	[1.00, 1.00]
School level
Average education				−0.32	0.72	[0.53, 0.98]
Average prestige				0.01	1.00	[0.98, 1.03]
Average income				−0.01	1.00	[1.00, 1.00]
Random effects		σ²	χ²		σ²	χ²
Intercept		1.16*	2,835.21		1.15*	2,775.11

Note. The alpha level was adjusted by the number of parameters estimated in the final model (i.e., α/n_p, or .05/13). AOR = adjusted odds ratio; CI = 95% confidence interval.

p < .005.

High-Incidence Disability Identification

The need for a hierarchical model was established by fitting a fully unconditional model. There was a significant amount of variation between schools in high-incidence disability identification (χ² = 1872.88, df = 1521, p < .001), indicating that a hierarchical model was appropriate for modeling variation in student identification. Following the unconditional model, the first model included student sex and race. As shown in Table 4, girls were significantly less likely to be identified as having a high-incidence disability (AOR = 0.37). Student race did not predict disproportionality in identification. The second model included student- and school-level SES predictors. None of the Level 2 variables significantly predicted student identification of having a high-incidence disability. Students whose parents had attained a higher education level were less likely to be identified as having a high-incidence disability (AOR = 0.73). Girls were still less likely to be identified with a high-incidence disability than boys (AOR = 0.36). Hierarchical linear modeling (HLM) does not provide residuals for binary dependent variables; therefore, the normality of Level 2 residuals is unknown. Because of the error in the odds ratio estimates, we cannot interpret the change in odds ratios across models.

Table 4.

Odds of Being Identified With a High-Incidence Disability.

	Model 1			Model 2
	β	AOR	CI	β	AOR	CI
Intercept	−2.81	0.06	[0.05, 0.08]	−2.93	0.05	[0.01, 0.24]
Student level^a
Girls	−1.01*	0.37	[0.26, 0.52]	−1.01*	0.35	[0.26, 0.52]
Black	0.23	1.26	[0.78, 2.05]	−0.01	1.00	[0.57, 1.74]
Hispanic	−0.01	1.01	[0.58, 1.75]	−0.34	0.71	[0.42, 1.20]
Education				−0.31*	0.73	[0.60, 0.91]
Prestige				−0.01	0.99	[0.97, 1.01]
Income				−0.01	1.00	[1.00, 1.00]
School level
Average education				−0.44	0.65	[0.42, 0.99]
Average prestige				0.01	1.00	[0.97, 1.04]
Average income				−0.01	1.00	[1.00, 1.00]
Random effects		σ²	χ²		σ²	χ²
Intercept		0.85*	1,862.71		0.90*	1,819.41

Note. The alpha level was adjusted by the number of parameters estimated in the final model (i.e., α/n_p, or .05/11). AOR = adjusted odds ratio; CI = 95% confidence interval.

Fewer than 10 Asian Americans had high-incidence disabilities and had to be excluded from these analyses.

p < .005.

Discussion

The purpose of this study was to investigate the relationship between tripartite measures of SES and racial disproportionality in special education. Previous research has offered conflicting findings on the relations of SES to disproportionality, with few studies investigating both student and aggregate measures concurrently (for exceptions, see Hibel et al., 2010; Sullivan & Bal, 2013), and fewer still offering conceptual or theoretical rationales for variable selection. Our measures—parent income, educational attainment, and occupational prestige—were based on the tripartite model of SES and have been shown elsewhere to be differentially related to health and education outcomes. Consistent with previous research in health and education (APA, 2007; Sirin, 2005; Veensra, 2000), the components of SES were differentially related to disproportionality. Our results showed that parents’ income and prestige were unrelated to special education identification, but higher levels of parent education were related to reduced risk of identification with a high-incidence disability. In addition, with the covariate included in our models, there was little evidence of racial disproportionality; only Asian American students were significantly underrepresented.

Our results underscore the importance of SES variable selection in the study of disproportionality. Findings from previous studies have varied, offering contradictory depictions of the relations of SES to special education identification, but any synthesis across studies is prevented by the range of measures used. Here, we selected three distinct measures of SES at the child and school levels based on the tripartite model of SES, and found that only parent education predicted identification in the high-incidence categories. The school measures were not significant, as has been shown in other recent large-scale analyses (Hibel et al., 2010; Sullivan & Bal, 2013). Taken together, this may indicate that efforts to understand the potential relations of SES should focus on the student and their family, and the results to school interactions and experience, rather than inferring from group or building composites. This offers a sharp contrast to early disproportionality studies where researchers investigated disparities in school and district identification rates relative to district or community SES measures. Taken together, our results along with those of other multilevel studies (e.g., Sullivan & Bal, 2013) suggest that patterns of individual risk are not comparable with aggregate measures of identification previously examined in studies of data from the Office of Special Education (OSEP), Office of Civil Rights, and state data systems. This distinction based on unit of analysis may help to elucidate the current debate surrounding increasing divergence characterizing this body of literature (Skiba et al., 2016). More research is needed to understand the implication of unit of analysis for understanding of patterns and correlates of disproportionality.

Unlike other recent findings of significant underrepresentation of racial minority students in special education and various disabilities (Hibel et al., 2010; Morgan et al., 2015), our findings do not suggest disproportionality in special education identification except for Asian American students, at least not in third grade. This may be due to our selection of distinct measures of SES following recommendations from the literature on the conceptual models of SES and the relative importance of each dimension in previous research (APA, 2007; Sirin, 2005), rather than the composite scores used elsewhere (Hibel et al., 2010; Morgan et al., 2015). If an aggregate measure of SES had been used, the lack of relationship of prestige and income to identification may have reduced the overall impact of SES. Likewise, considering only income as a covariate would have indicated that SES is unrelated to special education identification. This suggests that the common reliance on FRL may be inappropriate, which together with cogent critiques of the measure (Harwell & LeBeau, 2010; Lubienski, & Crane, 2010; Sirin, 2005) suggests a need to focus on other dimensions of SES. Notably, our results point to the importance of parent education in identification of high-incidence disabilities, while suggesting that differences in parents’ income and prestige do not contribute to differential decisions or treatment. Particularly where researchers have limited resources to collect SES information or seek to build parsimonious models, it may be wise to use parent education as a measure of SES.

In particular, these findings suggest that higher levels of parental educational attainment predict lower risk of high-incidence disabilities. Future research should examine the mechanisms by which parent education affects risk of special education identification, as there are several potential explanations to be explored. It is possible that higher parental education, particularly maternal education, reduces risk of developing mild disabilities (Mollborn, Lawrence, James-Hawkins, & Fomby, 2014), but the relations observed here may also indicate educational attainment influences parents’ educational decisions or interactions with schools. Perhaps parents with higher level of education avoid what they perceive as pejorative disability labels or unattractive services in favor of alternate educational supports, or advocate for their children in different ways in the team processes that often characterize referral and eligibility determination. Qualitative investigations may shed light on the dynamics of parent-school interactions that influence identification for special education or alternative pathways to academic, behavioral, or social-emotional supports outside of special education where students demonstrate educational difficulties that might otherwise result in special education services.

We have far to go before we fully understand the determinants of disproportionality. Even in a literature characterized by increasing discordance rather than convergence, it appears assumptions of race-based risk are unfounded. Given the social construction of race, authors should also avoid essentialization of race/ethnicity (Chao, Hong, & Chiu, 2013) in the study of disproportionality and instead consider how race influences students’ and families’ interactions with schools and may contribute to differences in special education identification. Indeed, race has no genetic basis (Yudell, Roberts, DeSalle, & Tishkoff, 2016), and we as a field still have limited understanding or ability to identify “true disability,” so to infer that race confers risk of disability in and of itself is erroneous. Instead, we should continue to explore how race and class relations intersect to shape students’ educational experiences and perceptions of special needs (e.g., Artiles, 2013; Chhuon & Sullivan, 2013).

Given the contradictions that increasingly characterize the disproportionality literature, it is premature to posit policy or practice implications even though our results differ from many previous findings. Our results may accurately reflect the dynamics under study here, but they may also be artifacts of the sample, instrumentation, measures, or something else. These findings should be replicated in other samples and settings. Although recent studies run counter to much of the earlier disproportionality studies, they often come from the same large-scale datasets or rely on the same instrumentation used in multiple federal data collection efforts. We must critically evaluate the samples and instruments used in any study, and strive to implement rigorous, theoretically based studies using conceptually sound measures to extend our understanding of disproportionality.

Limitations and Future Directions

Although rigorous methodology was used in our study, limitations of our findings should be noted. Although the ECLS-K allowed for analysis of both student/family and school predictors of special education risk in a large nationally representative sample, the sample was not without limitations. In particular, it did not allow for analysis of special education risk among American Indians or by disability category, so our analyses do not fully represent potential disparities in special education. The sample also did not allow for consideration of nuanced geographical or contextual differences in disproportionality which may be important as some research indicates disparities are related to district characteristics (e.g., Finn, 1982) or local enactment of policy (J. Oakes, Welner, Yonezawa, & Allen, 2005), which may be masked in national samples such as the ECLS-K.

In addition, these analyses were restricted to third-grade identification for special education. Later identified students, as well as those who may have exited special education prior to third grade, may not have been captured in this sample. Future research should examine the longitudinal relations of SES and race to special education outcomes. Disability identification was based on special education teacher reports subject to a high nonresponse rate and may have resulted in underreporting of categories, consistent with comparisons of special education rates in the ECLS-K relative to OSEP rates reported elsewhere (Skiba et al., 2016). This limitation could be improved upon by relying more heavily on records rather than educator reports.

We did not control for academic or behavioral performance given longstanding contention and challenges—both empirical and conceptual—to their limited utility in identification or differentiation of special needs, and their common correlation (e.g., Gage, Lierheimer, & Goran, 2012; Sabornie, Cullinan, Osborne, & Brock, 2005). Although this decision may be controversial, research indicates that a variety of contextual factors influence both educational opportunity and performance (e.g., Ferreira & Gignoux, 2011; Duncan & Murnane, 2011) and that teacher bias can affect student performance and measurement error (Staat & Patton, 2013; Mason, Gunersel, & Ney 2014). Taken together, these bodies of research challenge the utility of performance measures in the study of disproportionality because the research suggests low performance may be a consequence of interpersonal and systemic bias rather than special needs.

Finally, the ECLS-K was conducted between 1998 and 2007. While it provides a valuable counterpoint to much of the early disproportionality research (see Waitoller, Artiles, & Cheney, 2010, for description), it may not provide an accurate representation of the relations between SES and disproportionality today. Researchers should attempt to replicate these findings in other samples, particularly given changes in the Individuals with Disabilities Education Act (20 U.S.C. § 1400, 2004) regarding monitoring of disproportionality with the potential to systematically influence educational practice and decisions. Nonetheless, this study makes a valuable contribution to this literature because it examines the independent relations of multiple dimensions of SES at both the student and school levels using a large, nationally representative data set.

Conclusion

The present analysis supports the value of using componential measures of SES in the study of disproportionality because the various dimensions of SES are differentially related to outcomes. These findings suggest that, with the exception of parent education, student- and school-level socioeconomic factors do not provide additional predictive information when examining risk of special education identification. This finding is distinct from much of the previous literature that relied on district aggregates, FRL, or composite measures of SES, and highlights the need for further attention to the ways parent education influences students’ educational experiences and practitioners’ educational decisions. The most striking result here, however, was the lack of disparities for Black and Hispanic students, and should be replicated in other rigorous investigations. These findings also underscore the need for consideration of the conceptual and methodological nuances of this literature, especially in light of increasingly divergent and controversial findings and the varied methodological limitations and atheoretical investigations that characterize this body of work. We still have far to go before we understand the mechanisms through which disproportionality emerges or how to effectively address it. Given persistent concerns for inappropriate identification and statutory requirements to eliminate minority disproportionality in special education a la requirements for monitoring and early intervening in Individuals With Disabilities Education Act (IDEA), more research is needed to clarify the relations of race, SES, and disability. Such scholarship has the potential to inform reform and policy efforts at district, state, and federal levels.

Footnotes

Associate Editor: Audrey Trainor

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.

References

Albrecht

S. F.

Skiba

R. J.

Losen

D. J.

Chung

Middelberg

(2012). Federal policy on disproportionality in special education: Is it moving us forward? Journal of Disability Policy Studies, 23, 14–25.

American Psychological Association. (2007). Report of the APA task force on socioeconomic status. Washington, DC: Author.

Artiles

A. J.

(1998). The dilemma of difference: Enriching the disproportionality discourse with theory and context. The Journal of Special Education, 32, 32–36.

Artiles

A. J.

(2013). Untangling the racialization of disabilities: An intersectionality critique across disability models. DuBois Review, 10, 329–347.

Blair

Scott

K. G.

(2002). Proportion of LD placements associated with low socioeconomic status: Evidence for a gradient? Journal of Special Education, 36, 14–22.

Bollen

K. A.

Glanville

J. L.

Stecklov

(2001). Socioeconomic status and class in studies of fertility and health in developing countries. Annual Review of Sociology, 27, 153–185.

Bollmer

Bethel

Garrison-Mogren

Brauen

(2007). Using the risk ratio to assess racial/ethnic disproportionality in special education at the school-district level. Journal of Special Education, 41, 186–198. doi:10.1177/00224669070410030401

Chao

M. M.

Hong

Chiu

(2013). Essentializing race: Its implications on racial categorization. Journal of Personality and Social Psychology, 104, 619–634.

Chhuon

Sullivan

A. L.

(2013). Racialization of abilities and disabilities in U.S. schools: Asian American and Pacific Islander students in gifted and special education. Perspectives on Communication Disorders and Sciences in Culturally and Linguistically Diverse Populations, 20, 49–59.

10.

Cohen

D. R.

Burns

M. K.

Riley-Tillman

Hosp

(2015). Are minority students under-or overrepresented in special education? Communiqué, 44(2), 1, 22–23.

11.

Coutinho

M. J.

Oswald

D. P.

Best

A. M.

(2002). The influence of sociodemographics and gender on the disproportionate identification of minority students as having learning disabilities. Remedial and Special Education, 23, 49–59.

12.

Donovan

M. S.

Cross

C. T.

(Eds.). (2002). Minority students in special and gifted education. Washington, DC: National Academy of Science.

13.

Duncan

G. J.

Murnane

R. J.

(Eds.). (2011). Whither opportunity: Rising inequality, schools, and children’s life chances. New York, NY: Russell Sage.

14.

Dunn

L. M.

(1968). Special education for the mildly mentally retarded: Is much of it justifiable? Exceptional Children, 35, 5–22.

15.

Eitle

T. M.

(2002). Special education or racial segregation: Understanding variation in the representation of Black students in educable mentally handicapped programs. Sociological Quarterly, 43, 575–605.

16.

Ferreira

F. H. G.

Gignoux

(2011). The measurement of educational inequality: Achievement and opportunity [Discussion paper series, IZA DP No. 6161]. Retrieved from http://ftp.iza.org/dp6161.pdf

17.

Finn

J. D.

(1982). Patterns in special education placement as revealed by the OCR surveys. In Heller

K. A.

Holtzman

W. H.

Messick

(Eds.), Placing children in special education: A strategy for equity (pp. 322–381). Washington, DC: National Academy Press.

18.

Gage

N. A.

Lierheimer

K. S.

Goran

L. G.

(2012). Characteristics of students with high-incidence disabilities broadly defined. Journal of Disability Policy Studies, 23, 168–178.

19.

Gresham

F. M.

Sugai

Horner

R. H.

(2001). Interpreting outcomes of social skills training for students with high-incidence disabilities. Exceptional Children, 67, 331–344.

20.

Harwell

LeBeau

(2010). Student eligibility for a free lunch as an SES measure in education research. Educational Researcher, 39, 120–131. doi:10.3102/0013189X10362578

21.

Hibel

Farkas

Morgan

P. L.

(2010). Who is placed into special education? Sociology of Education, 83, 312–332.

22.

Hibel

Jasper

A. D.

(2012). Delayed special education placement for learning disabilities among children of immigrants. Social Focus, 91, 503–530.

23.

Hosp

J. L.

Reschly

D. J.

(2004). Disproportionate representation of minority students in special education: Academic, demographic, and economic predictors. Exceptional Children, 70, 185–199.

24.

Individuals with Disabilities Education Act of 2004, 20 U.S.C. § 1400 (2004).

25.

Klingner

J. K.

Artiles

A. J.

Kozleski

Harry

Zion

Tate

. . . Riley

(2005). Addressing the disproportionate representation of culturally and linguistically diverse students in special education through culturally responsive educational systems. Education Policy Analysis Archives, 13. Retrieved from http://epaa.asu.edu/ojs/article/view/143

26.

Lubienski

S. T.

Crane

C. C.

(2010). Beyond free lunch: Which family background measures matter. Education Policy Analysis Archives, 18(11), 1–39. Retrieved from http://epaa.asu.edu/ojs/article/view/756

27.

MacMillan

D. L.

Reschly

D. J.

(1998). Overrepresentation of minority students: The case for greater specificity or reconsideration of the variables examined. The Journal of Special Education, 32, 15–24.

28.

Marx

Engels

(2004). Class struggle. In Lemert

(Ed.), Social theory: The multicultural and classic reading (3rd ed., pp. 37–41). Boulder, CO: Westview Press. (Original work published 1848)

29.

Mason

B. A.

Gunersel

A. B.

Ney

E. A.

(2014). Cultural and ethnic bias in teacher ratings of behavior: A criterion-focused review. Psychology in the Schools, 51, 1017–1030.

30.

Mollborn

Lawrence

James-Hawkins

Fomby

(2014). When do socioeconomic resources matter most in early childhood? Advances in Life Course Research, 20, 56–69.

31.

Morgan

P. L.

Farkas

Hillemeier

M. M.

Mattison

Maczuga

Cook

(2015). Minorities are disproportionately underrepresented in special education: Longitudinal evidence across five disability conditions. Educational Researcher, 44, 278–292.

32.

Morgan

P. L.

Staff

Hillemeier

M. M.

Farkas

Maczuga

(2013). Racial and ethnic disparities in ADHD diagnosis from kindergarten to eighth grade. Pediatrics, 132, 85–93.

33.

Nakao

Treas

(1990). Computing 1989 occupational prestige scores (No. 70). Chicago, IL: National Opinion Research Center.

34.

National Center on Education Statistics. (2012). The condition of education 2012. Washington, DC: U.S. Department of Education.

35.

Oakes

J. M.

Rossi

P. H.

(2003). The measurement of SES in health research: Current practice and steps toward a new approach. Social Science & Medicine, 56, 769–784.

36.

Oakes

Welner

Yonezawa

Allen

R. L.

(2005). Norms and politics of equity-minded change: Researching the “zone of mediation.” In Fullan

(Ed.), Fundamental change: International handbook of educational change (pp. 282–305). Dordrecht, The Netherlands: Springer.

37.

Oswald

D. P.

Coutinho

M. J.

Best

A. M.

(2002). Community and social predictors of overrepresentation of minority children in special education. In Losen

D. J.

Orfield

(Eds.), Racial inequity in special education (pp. 1–13). Cambridge, MA: Harvard Education Press.

38.

Oswald

D. P.

Coutinho

M. J.

Best

A. M.

Nguyen

(2001). Impact of the sociodemographic characteristics on the identification rates of minority students as having mental retardation. Mental Retardation, 39, 351–367.

39.

Oswald

D. P.

Coutinho

M. J.

Best

A. M.

Singh

N. N.

(1999). Ethnic representation in special education: The influence of school-related economic and demographic variables. The Journal of Special Education, 32, 194–206.

40.

Raudenbush

S. W.

(2011). Optimal design software for multi-level and longitudinal research (Version 3.01) [Software]. Available from www.wtgrantfoundation.org

41.

Ridgeway

C. L.

(2014). Why status matters for inequality. American Sociological Review, 79, 1–16.

42.

Sabornie

E. J.

Cullinan

Osborne

S. S.

Brock

L. B.

(2005). Intellectual, academic, and behavioral functioning of students with high-functioning disabilities: A cross-categorical meta-analysis. Exceptional Children, 72, 47–63.

43.

Shifrer

Muller

Callahan

(2010). Disproportionality and learning disabilities: Parsing apart race, socioeconomic status, and language. Journal of Learning Disabilities, 44, 246–257.

44.

Sirin

S. R.

(2005). Socioeconomic status and academic achievement: A meta-analytic review of research. Review of Educational Research, 75, 417–453.

45.

Skiba

R. J.

Artiles

A. J.

Kozleski

E. B.

Losen

D. J.

Harry

E. G.

(2016). Risk and consequences of oversimplifying educational inequities: A response to Morgan et al. (2015). Educational Researcher, 45, 221–225.

46.

Skiba

R. J.

Poloni-Staudinger

Simmons

A. B.

Feggins-Azziz

Chung

(2005). Unproven links: Can poverty explain ethnic disproportionality in special education? Journal of Special Education, 39, 130–144.

47.

Snijders

T. A. B.

Bosker

R. J.

(1999). Multilevel analysis: An introduction to basic and advanced multilevel modeling. Thousand Oaks, CA: SAGE.

48.

Staat

Patton

(2013). State of the science: Implicit bias review 2013. Columbus, OH: Irwan Institute. Retrieved from http://kirwaninstitute.osu.edu/docs/SOTS-Implicit_Bias.pdf

49.

Sullivan

A. L.

Artiles

A. J.

(2011). Theorizing racial inequity in special education: Applying structural theory to disproportionality. Urban Education, 46, 1526–1552.

50.

Sullivan

A. L.

Bal

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

51.

Tourangeau

Nord

Sorongon

A. G.

(2009). Early Childhood Longitudinal Study Kindergarten Class of 1998-99 (ECLS-K), Combined user’s manual for the ECLS-K Eighth-Grade and K–8 full sample data files and electronic codebooks (NCES 2009-004). Washington, DC: U.S. Department of Education.

52.

Tourangeau

Nord

Wan

(2002). Early Childhood Longitudinal Study Kindergarten Class of 1998-99 (ECLS-K), user’s guide to the longitudinal kindergarten-first grade public-use data file (NCES 2002-149). Washington, DC: U.S. Department of Education.

53.

Veensra

(2000). Social capital, SES, and health: An individual-level analysis. Social Science & Medicine, 50, 619–629.

54.

Waitoller

F. R.

Artiles

A. J.

Cheney

D. A.

(2010). The miner’s canary: A review of overrepresentation research and explanations. Journal of Special Education, 44, 29–49. doi:10.1177/0022466908329226

55.

Weber

(2004). Class, status, party. In Lemert

(Ed.), Social theory: The multicultural and classic readings (3rd ed., pp. 115–126). Boulder, CO: Westview Press. (Original work published 1920)

56.

Yudell

Roberts

DeSalle

Tishkoff

(2016). Taking race out of human genetics: Engaging a century-long debate about the role of race in science. Science, 351, 564–565.