The purpose of this study was to investigate the extent of gender and race/ethnicity disproportionality among students identified as having learning disabilities (LD) and to investigate relationships between disproportionality and sociodemographic factors. Using nationally representative data collected by the U.S. Office for Civil Rights, this study examined the effects of gender, ethnicity, and sociodemographic factors on the proportion of students who are identified as having LD. Results indicated a clear association between ethnicity and gender and the odds of being identified as a student with LD. Sociodemographic factors for a school district were also found to be strongly associated with the proportion of students identified as having LD. A logistic regression model that included the nine sociodemographic variables, gender, and race, was significantly better at predicting LD identification than a model that included sociodemographic predictors alone. Adjusted odds ratios illustrated how the likelihood of identifying LD changes when sociodemographic influences are taken into account. Findings indicated that both individual student characteristics and district sociodemographic characteristics are important in determining the likelihood of LD identification and that the impact of the sociodemographic characteristics is different for each of the gender-ethnicity groups.
Concern about disproportionate representation of minority students in special education persists despite more than 30 years of litigation, debate, and initiatives (Artiles, Aguirre-Munoz, & Abedi, 1998; Dunn, 1968; Heller, Holtzman, & Messick, 1982; Larry P. v. Wilson Riles, 1972, 1979, 1984, 1986). Most often, the focus has been on students who are African American and/or those with high-incidence disabilities, although recently some interest has been generated in examining the significance of disproportionality for other race/ethnicity groups and among students with severe disabilities (Oswald & Coutinho, 2001; US Department of Education, 2000).
The largest proportion of students in special education continues to be those identified as having learning disabilities (LD); therefore, any overrepresentation among students with LD must be regarded as nationally significant (US Department of Education, 2000). From 1989 to 1999, the number of students placed in the LD category increased by 36.6%. This is the largest percentage change among the high-incidence disability conditions (i.e., mental retardation [MR], speech and language impairments, and serious emotional disturbance [SED]). During the 1998-1999 school year, approximately 2.8 million 6- to 21-year-old children were identified as having LD, representing slightly more than 50% of all students who were reported as having disabilities under the Individuals with Disabilities Education Act (IDEA) of 1990 (US Department of Education, 2000).
Evidence for disproportionate representation of minority students among students with LD has been reported, although less often than for mental retardation and emotional disturbance (Anderson & Webb-Johnson, 1995; Coutinho, Oswald, Best, & Forness, in press; Gregory, Shanahan, & Walberg, 1986; Harry, 1992; Harry & Anderson, 1994; Lara, 1994; Ortiz & Yates, 1983; Robertson, Kushner, Starks, & Drescher, 1994). In 1994, Robertson et al. examined the percentage of students with LD, by ethnic category, in 15 cities and reported that compared to other ethnic groups, a higher percentage of African American students were identified as having LD in 10 of the 15 cities. Hispanic students were underrepresented among students with LD in 10 of the 15 cities. The percentage of American Indian students with LD was not reported because of the small number of American Indian students in the cities studied (see Note).
Recently, an analysis of 1994 US Office for Civil Rights (OCR) data on students with LD found that African American and Hispanic students were identified at the same rate as White students, and American Indian students were 1.2 times as likely as students in the other group to be identified. Descriptive analyses of the 14-year time period revealed a persistent pattern of overrepresentation for American Indian students with LD. During the same time period, American Indian children were similarly overrepresented among students with mental retardation and students with serious emotional disturbance (Oswald & Coutinho, 2001).
The 1997 IDEA Amendments mandated that states collect special education “child count” data by race/ethnicity, beginning with the 1998-1999 school year. The US Department of Education (2000) reported special education identification rates by race/ethnicity and disability for children ages 6 through 21. According to this report, 4.27% of White children were identified as having LD; corresponding figures for other race/ethnicity groups were as follows: American Indian/Alaska native, 6.29%; Black, 5.67%; Hispanic, 4.97%; and Asian/Pacific Islander, 1.7%.
These studies underscore the importance of disaggregating special education identification data by race/ethnicity as well as by disability. Specific geographic regions may exhibit patterns that are markedly distinct from the national profile (Oswald & Coutinho, 2001; Robertson et al., 1994). Nonetheless, the data raise concern that race/ethnicity disproportionality among students with LD is on the rise.
Understanding why disproportionate representation occurs and deciding how to respond appropriately require both a coherent conceptual framework and meticulous empirical investigation (Utley & Obiakor, 2001). Artiles, et al. (1998) described the persistence of disproportionality along a continuum ranging from discriminatory professional practices to innate deficits of minority children. They recognized that both problematic eligibility practices and sociopolitical factors, such as school violence and school disciplinary practices, may influence disproportionality. Artiles et al. (1998) investigated whether family variables and student perspectives (sociocultural factors) predicted placement patterns for Hispanic, White, and African American students. Discriminant function analyses identified some LD placement predictors that differed across ethnic groups and some predictors that were significant for all students. For example, Hispanic students with LD (as compared to Hispanic students without LD [NLD]) demonstrated lower math achievement, less family structure, and smaller families. Significant predictors of LD identification for African American students were smaller families, a higher perception of social status, and a lower level of family structure/rules. Among factors that did not predict placement within any ethnic group were student self-esteem, locus of control, behavioral history, perception of school risk and protective factors, and perception of parent expectations. Interestingly, “African-American and Anglo students with LD reported a higher perception of social status than their NLD peers” (Artiles et al., 1998, p. 555). Student perception of academic standing was a significant predictor of LD identification only for White students.
Coutinho and Oswald (2000) offered two hypotheses regarding the causes of ethnic disproportionality in special education: the processes that measure and interpret the ability, achievement, and behavior of students (i.e., referral, assessment, eligibility) may work differently across ethnic groups, leading to disproportionality as an artifact of faulty assessment; and the underlying distribution of educational disability may vary across ethnic groups as a result of social and demographic influences that represent risk factors for disabilities.
The first hypothesis reflects concern that public education embodies cultural biases that incorrectly and disproportionately target minority students during the referral, assessment, and eligibility process. This hypothesis conceptualizes disproportionate representation as a sociopolitical, historical problem where teacher-student and home-school discontinuities and conflicts lead to the overrepresentation of minority students, particularly Black students, as having LD (Harry, Rueda, & Kalyanpur, 1999; Trent & Artiles, 1995; Utley & Obiakor, 2001). Variations among states in the definition and implementation of LD and other disability conditions, for example, represents a potential source of bias during the referral and eligibility process.
The second hypothesis is that minority groups may be differentially susceptible to educational disability. Social and demographic factors to consider include poverty, school and community fiscal factors, and access to appropriate general education options (Fujiura & Yamaki, 2000; Messick, 1984; US Department of Education, 1998; Utley & Obiakor, 2001). Oswald, Coutinho, Best, and Singh (1999) reported that a set of community- and school-related variables accounted for a significant portion of the variability in school districts’ identification rates of African American students as having mental retardation or SED. More recently, Oswald, Coutinho, Best, and Nguyen (in press) reported further analyses of the 11994 OCR data for students with mental retardation.
Purposes of the study
The purpose of the present study was to improve understanding of the influence of individual student and district-level variables that influence special education identification rates for minority students. Earlier studies examined a variety of sociopolitical and sociodemographic predictors of LD but did not include results by ethnicity (Coutinho & Oswald, 1998; Lester & Kelman, 1997). The current study investigated the extent of disproportionality among students with LD and sought to describe the relationship between disproportionality and sociodemographic variables. Thus, we included sociodemographic predictors in a model constructed to describe the variability in LD prevalence among communities.
Approximately every 2 years, the US Department of Education OCR collects information on a nationally representative sample of school districts. This survey is the chief source of data on the status of civil rights in the nation’s schools. Approximately one third of the nation’s school districts are included in the stratified random sample (US Department of Education, 1998), and state and national figures may be projected from the survey data.
For this study, we considered only the information on enrollment and disability categories from the school year 1994-1995. District enrollment was broken down by ethnicity (five categories) and by gender (two categories). The number of students in three disability categories (MR, SED, and LD) was also broken down by ethnicity and by gender. The fourth disability category in this study, “None,” included students with low-incidence disability conditions as well as all students who were not identified as having disabilities. This article focuses only on findings related to LD identification.
Sample weights were used in all analyses to reflect the fact that all of the nation’s large school districts are included in the data set but only a few small districts are included. The weighted number of students in each disability category is provided in Table 1. Districts that were “forced” into the sample-usually for monitoring reasons-carried weights of zero and were excluded from the analyses reported here.
The original data file had 44,276 observations (i.e., schools) and 431 variables. We collapsed the data by accumulating the data for all the schools in a district because the predictor variables used in the study were available only at the district level. Included in the sample were 4,151 school districts that served more than 24 million students.
The National Center for Educational Statistics (NCES) Common Core of Data CD-ROM (NCES, 1993) includes information on 15,041 school districts in the 50 states and Washington, DC. For this study, the NCES data set was merged with the OCR data so that only districts included in the 1994 OCR sample were retained. In the process of merging the two data sets, we discovered about 30 districts for which the ID numbers were inconsistent. Through a process of comparing the districts’ names, cities, and states, we were able to correct all but two of these inconsistencies; the two mismatches were dropped from the sample.
Nine sociodemographic variables were chosen from the NCES data set as predictor variables for this study: student teacher ratio (STR), per-pupil expenditure (PPE), percentage of children enrolled who were considered at risk (At Risk), percentage of enrolled students who were non-White (Non-white), percentage of enrolled students who were limited English proficient (LEP), median housing value for houses, in $10,000 units (Housing), median income for households with children, in 100,000 units (Income), percentage of children in households below poverty level (Poverty), and percentage of adults in the community who had education of 12th grade or less and no diploma (No Diploma). Additional variables were considered but excluded due to a substantial number of missing values (e.g., districts’ dropout rates). In a few cases, a district’s information was not available for the most recent school year, and a previous year’s data was used to fill out the data set. As Table 2 shows, the variable with the most missing values, STR, was available for 95.2% of the districts and 94.5% of all students in the sample.
The predictor variables included in the analyses were clearly not unrelated; Spearman rank-order correlation coefficients for predictor variables are provided in Table 3. Poverty and Income, for example, showed a predictable strong inverse relationship.
We examined the effects of gender, ethnicity, and sociodemographic factors on the proportion of students in a school district identified as having LD. The models used the proportion of students in the LD category as the dependent variable and two sets of variables as predictors. The predictor variables were the district-level sociodemographic continuous variables and the child-level categorical variables of gender and ethnicity. Categorical responses (such as Learning Disability: Yes/No) can be modeled by using logistic regression (Hosmer & Lemeshow, 1989), and such models include the proportion in each disability category as the response variable. The logistic regression procedure reported in this article models the probability of a child being classified as having LD as a function of gender, ethnicity, and sociodemographic predictors.
The model used in this study examined the probability of a child being in one of the four disability conditions; however, in the interest of clarity, we report only the LD findings in this article. All analyses were done with the districts weighted by the number of students (as well as by sample weight) so that the models simulate using the student rather than the district as the unit of analysis.
The predictor variables are of two types. The categorical predictor variables were gender (female, male) and ethnicity (American Indian, Asian/Pacific Islander, Black, Hispanic, and White). For the continuous sociodemographic covariates listed in Table 2, effects in the model included a linear and quadratic trend for each covariate, all possible two-way interactions for the linear and quadratic trends for each covariate, and gender, ethnicity, and Gender x Ethnicity interaction effects crossed with the linear and quadratic trends for each covariate. The net effect of this model is the possibility of a separate linear and quadratic trend for each gender and ethnicity combination. All continuous covariates were centered and scaled to avoid problems with ill conditioning and collinearity in analyses (Draper & Smith, 1998). Because of the large sample size and the complexity of the model, p < .0005 was used as the cut-off for significance.
For the sample as a whole, 5.5% of all students were identified as having LD. This was comparable to the 1994-1995 LD identification rate of 5.7% of enrolled students, reported in the 1996 Annual Report to Congress (US Department of Education, 1996). Table 4 illustrates, however, that the LD identification rate for gender/ethnicity groups varied widely from a low of 1.2% for Asian/Pacific Islander females to a high of 9.2% for American Indian males, clearly demonstrating disproportionate identification across the groups.
A simple chi-square analysis of the data in Table 1 shows a significant association between ethnicity and gender and the disabling conditions, X2(27, N = 41,819,191) = 628,912, p < .0001, verifying that the LD identification rates are not the same for all 10 gender/ethnicity groups. To clarify this finding, we constructed odds ratios for each of the gender/ethnicity groups, with White females as the comparison group. These odds ratios provide an estimate of the likelihood of being identified as having LD compared to the likelihood for White female students. Thus, the odds ratio for White female students is, by definition, 1.0. In this sample, White males were 2.3 times as likely as White females to be identified as having LD, and Black females were 0.9 times as likely (see Table 4). American Indian males display ‘the largest disproportionality, with an odds ratio of 2.9.
A logistic regression analysis with LD identification as the response variable and only the nine sociodemographic variables as predictors (including linear, quadratic, and interaction effects) was significant, X 2(162, N = 41,819,191) = 345,130, p < .0001. Thus, the sociodemographic conditions of a school district are strongly associated with the proportion of students identified as having LD; some statistically significant portion of the variation in districts’ LD identification rates can be explained by this combination of predictor variables.
The bivariate correlation coefficients between identification rate for each gender/ethnicity group and the predictors are provided in Table 5. These descriptive statistics illustrate the general direction and strength of the relationship between each predictor variable and LD identification rates. Spearman rank correlations were used for this purpose because of the skewed distributions of the variables and the presence of marked outliers. These correlation coefficients should be interpreted with caution because the relationships between identification rates and predictors are not linear, as demonstrated by the significant quadratic and interaction effects in the logistic model. Further, because these are bivariate correlations, the relationships do not take into account the effects of other predictors in the model.
The overall LD identification rate was positively associated most strongly with the variables non-White students and LEP, but even these relationships were relatively weak. Further, relationships between LD identification and predictor variables for individual gender/ethnicity groups were mixed in terms of strength and direction. Housing and Income, for example, were weakly to moderately positively associated with LD identification for all groups except American Indian students. Nonwhite and LEP were weakly to moderately positively associated with LD identification for all groups except White students. PPE was weakly positively associated, except with American Indian and Black students. No Diploma was weakly to moderately negatively associated for Asian/Pacific Islander and Hispanic students and White female students. Other variables were mixed in terms of significance and the direction of the relationships.
Given that individual student characteristics (i.e., gender and ethnicity) and sociodemographic variables were each separately associated with the identification of students with LD, we next sought to determine whether these two classes of variables contributed uniquely to the prediction of LD identification and whether the relationships between the predictor variables and LD identification rates were the same for each gender/ethnicity group. A logistic regression analysis including the nine sociodemographic variables (and their linear, quadratic, and interaction effects), gender, race, and all possible interactions of the covariates with gender and race was found to be significantly better than the model with only the sociodemographic predictors, X2(1,485, N = 41,819,191 ) = 667,570, p < .0001, and significantly better than the model with only the gender and ethnicity groups, X2(1,620, N = 41,819,191) = 383,788, p < .0001 . There was also a significant gender/ethnicity-by-sociodemographic interaction, X 2(1,458, N = 41,819,191 ) = 86,224, p < .0001, indicating that both individual student characteristics and district sociodemographic characteristics are important in determining the likelihood of LD identification and that the impact of the sociodemographic characteristics is different for each of the various gender/ethnicity combinations.
To illustrate this finding, we plotted the predicted LD identification rate for each gender/ethnicity group across the range of poverty represented in the sample. For the purpose of this illustration, the effects of all other sociodemographic variables are held constant (i.e., set to the median value). Figure 1 shows that, for Black and Hispanic students, LD identification increases substantially as poverty increases, particularly for males. For American Indian and White students, identification tends to decrease with increasing poverty, again, particularly for males.
Conversely, when plotted across the full range of non-White, Figure 2 shows that for most gender/ethnicity groups, LD identification tends to decline as the non-White percentage in the district increases, dramatically so for Hispanic and Black males. For American Indian students, however, LD identification tends to increase slightly as the non-White percentage increases.
To clarify how the odds ratios for the gender/ethnicity groups change when sociodemographic influences are taken into account, we calculated adjusted odds ratios, by calculating the odds ratio for each gender/ethnicity group at the median value of each of the predictors. This odds ratio is actually a predicted value based on all the effects included in the logistic regression model. As shown in Table 4, the odds ratios change, but not necessarily in the expected direction. The adjusted odds ratios for Black and Hispanic students and for White male students actually increase slightly at the median value of the sociodemographic variables. Overrepresentation of American Indian students is diminished slightly when all the predictor variables are taken into account.
Finally, we examined how the odds ratios change across the distribution of each of the predictor variables. We computed the 10th percentile and the 90th percentile for each predictor and computed the odds ratio for each of the gender/ ethnicity groups at those points in the distribution, holding all the other predictors at the median value. The results are shown in Table 6.
For example, as Housing goes from the 10th percentile ($42,733) to the 90th percentile ($192,027), predicted disproportionality for Black and Hispanic male students increases dramatically. Interestingly, as LEP goes from the 10th percentile (0.2%) to the 90th percentile (6.4%), disproportionality among Hispanic students declines such that, in districts with the highest LEP rates, Hispanic students are identified as having LD at about the same rate as White students.
The data in Table 6 also illustrate how the predictors differentially affect the gender/ethnicity groups. The disproportionality of LD identification for Black male students, for example, increases substantially as Income goes from the 10th to the 90th percentile, whereas for American Indian male students, the pattern changes from marked overrepresentation to moderate underrepresentation.
FIGURE 1. Learning disabilities and poverty.
Note. M AI = male American Indian; M AS = male Asian/Pacific Islander; M BL = male Black; M HI = male Hispanic; M WH = male White; F AI female American Indian; F AS = female Asian/Pacific Islander; F BL = female Black; F HI = female Hispanic; F WH = female White.
FIGURE 2. Learning disabilities and non-White enrollment.
Note. M Al = male American Indian; M AS = male Asian/Pacific Islander; M BL = male Black; M HI = male Hispanic; M WH male White; F Al = female American Indian; F AS = female Asian/Pacific Islander; F BL = female Black; F HI = female Hispanic; F WH = female White.
The data illustrate that the problem of disproportionality in special education identification is not limited to African American students nor to the categories of MR and SED. These findings highlight a relatively neglected instance of disproportionality: the overrepresentation of American Indian males among students with LD.
The data also support the position that both individual student characteristics, such as gender and ethnicity, and communities’ sociodemographic characteristics influence the likelihood that a child will be identified as having LD. These findings offer some clues to clarify the validity of alternative hypotheses regarding the causes of disproportionality. Increased poverty, for example, is associated with increased LD identification rates among Black, Hispanic, and male Asian students. This result provides additional supportive evidence for the well-established finding that poverty is associated with increased risk for a variety of disability conditions, including LD. Thus, minority groups that experience more poverty than people categorized as White might be expected to have more LD (i.e., minority children may be differentially susceptible to LD because of higher poverty rates). Somewhat inexplicably, however, the effect is reversed for White and American Indian students; among these groups, increased poverty is associated with lower LD identification rates. Thus, the evidence for differential susceptibility is mixed, at best.
On the other hand, LD identification rates for all gender/ ethnicity groups (except American Indian students) decline as the proportion of nonwhite students in the district increases, and there is no clear conceptual explanation for an association between LD identification and the nonwhite percentage. Further, the fact that the direction of the association is reversed for the American Indian group raises the concern that the ethnic makeup of the community is somehow differentially affecting LD identification for this group of students. Such a finding offers indirect support for the hypothesis that the processes for identification work differently for different gender/ethnicity groups, suggesting the possibility of inadvertent or deliberate bias.
Implications for Practice
These findings support a general conclusion that disproportionality is multiply determined and underscore the importance of understanding the causes in a specific school district before proceeding to an intervention. Disproportionality among minority students identified as having LD has not been studied as frequently as for other disability conditions, notably MR and SED. Attention is needed to ensure that an appropriate response is implemented for students with LD - the group that makes up more than 50% of all students with disabilities. LD is often considered a less stigmatizing disability than other types of disabilities, and issues surrounding the appropriate identification and educational needs of students with LD could easily be overlooked or addressed inappropriately through the implementation of procedures that were developed with other disability conditions in mind (Kavale & Forness, 1995; Ortiz & Maldonado-Colon, 1986). Similarly, the findings reported here underscore the importance of advocacy and continued research focused on American Indian students, with particular attention to why the effects of some of the sociodemographic predictors for them are reversed.
Bias may be introduced at different points in the special education identification process. Teachers may incorrectly refer minority students who are not disabled but who behave, attend, or learn somewhat differently than White, middle class students. In these cases, differences in behavior or learning needs may be interpreted as disability rather than acknowledged as cultural difference.
Bias may also occur in the application of the exclusionary provision of the LD definition. According to this provision students whose learning or behavioral problems are due primarily to environmental causes (e.g., poverty or poor teaching) are not to be identified as LD. The results of the present study, however, indicate that Black and Hispanic students and male Asian students are more likely to be identified as having LD as poverty increases, raising the question of whether the exclusionary provision is being consistently applied. Educational assessment may not pinpoint the cause of LD, particularly in settings where a narrow application of the discrepancy model is employed. Alternatively, the exclusionary provision may be set aside deliberately if special education is perceived as the only way to help a student; nonetheless, such a failure to apply the provision may result in a biased eligibility process. Because more Black and Hispanic children than White children live in poverty, ignoring the provision results in a disproportionally large number of these children being identified as having LD.
Alternatively, the increased LD identification rates for Black and Hispanic students may reflect differential susceptibility to disability, in which case the exclusionary provision of the LD definition is either unworkable in practice or is untenable. If poverty and other social ills are in fact important factors in the etiology of LD, the provision may unfairly exclude children with genuine disability, and that unfair exclusion may be more common among non-White children than among White children.
Tension and uncertainty surround the contribution of poverty to disability, including LD, as is now apparent in the amended formula for federal funding of IDEA (1997 amendments to IDEA). That is, if identification rates rise above specified levels, federal monies are distributed disproportionately. At that point, funding is based on rates of poverty rather than only a flat per-child amount. Thus, there is some inconsistency in the federal code regarding the implied relationship between poverty and disability.
Appropriate responses to disproportionate representation at the community level involve several steps. First, objective and accurate data must be collected to determine the extent of disproportionality by ethnicity and gender. Odds ratios can be calculated for individual districts to get a snapshot of the extent of disproportionality; when viewed in conjunction with changes in identification rates over time, such odds ratios effectively characterize changes in disproportionality across ethnic and gender groups. Second, preservice and inservice training and support must be provided to teachers and administrators to ensure culturally competent teacher-student and school-home relationships. Third, the referral, assessment, and eligibility process must be monitored systematically for any evidence of bias. Direct observation of student behavior, typically regarded as an objective measure, may ignore teacher behavior or classroom variables that reflect a cultural discontinuity between a teacher and a minority student. Careful attention must be given to the process implemented by schools with low minority enrollments to ensure appropriate, nondiscriminatory identification practices. Fourth, monitoring is needed with regard to the application of the exclusionary provision of the definition of LD by Individualized Education Program (IEP) teams. Objective, reliable procedures for the application of the provision must be developed and implemented systematically. In addition, any evidence that placement in special education is an effort to compensate for poor-quality general education services must be addressed at the child and system levels. Finally, in addition to modifications in educational practices, evidence of disproportionate representation in communities with high rates of poverty may indicate a need for advocacy efforts to promote social changes that address disadvantaged populations.
Recommendations for Further Research
Further research is needed to determine how both individual student characteristics (e.g., gender, ethnicity, IQ, adaptive behavior) and community sociodemographic factors influence the likelihood that a minority student will be identified as having LD. Secondary analyses of national data provide the starting point for community-level follow-along studies to investigate both the possibility of bias and differential susceptibility to disability. Sites should be selected that are at risk on predictors such as poverty and minority percentage, which have been observed to be significantly related to disproportionality. Sites should also vary with respect to disproportional representation (i.e., some demonstrating high disproportionality and others with none or very low rates). Samples should include students who remain in or return to general education, those who are served through Chapter 1 programs, and those who are served through inclusive or pullout special education arrangements. Student and system information should be collected so that the special education eligibility process is examined for bias and so that student characteristics, educational placements, and outcomes are documented to test the hypothesis of differential susceptibility.
Appropriate responses to disproportionality, however, require consideration not only of the processes involved in special education identification but also consideration of information about education experiences and outcomes. For example, if the outcomes of minority students identified as having LD are positive (i.e., equal to or better than outcomes for other students with LD) and are better than outcomes for comparable minority students who are not identified, then even substantial disproportionality may be irrelevant because such outcomes demonstrate the effectiveness of special education interventions. If the outcomes for comparable identified and nonidentified minority students are both unsatisfactory, then the emphasis belongs not on disproportionality but on improving the educational experiences of all minority students, perhaps involving larger-scale social changes (Garcia & Malkin, 1993). If the outcomes for identified minority children are measurably inferior to the outcomes of similar, children who remain unidentified, the quality of the special education service is indicted, and disproportionality in such a system may indeed be discriminatory. Finally, if minority and majority student outcomes are unequal, even if there is no disproportionality, attention should shift to the quality of special education services. The special education experience may be differentially ineffective for minority students; even if these students are appropriately identified, if their outcomes are measurably inferior to those of majority students, it cannot be said that these students are receiving free, appropriate education according to IDEA. The interpretation of student outcomes in all such studies must be referenced to professionally accepted standards and the national commitment to improve the educational experience of minority students.
Effective responses to disproportionate representation-that is, responses that lead to improved educational experiences and outcomes for minority students-must be based on a clear understanding of how both bias and differential susceptibility to LD affect identification rates, educational experiences, and outcomes. Some of the predictor effects in the model support the hypothesis that disproportionality is, in part, a result of differential susceptibility. Environmental factors that influence disability prevalence are not uniformly distributed across ethnic groups, and a myopic determination to achieve equal identification may overlook the need for large-scale social change. Nonwhite children are disproportionately exposed to potentially toxic environmental influences, and this fact represents a critical bias in US society that cannot be ignored.
In addition, equality of opportunity and satisfactory outcomes for minority students require attention to equity (i.e., ensuring that services are both appropriate and sufficient; Epps, 2001). An education provided to Black, American Indian, or Hispanic students that is simply “equal” to that provided to White or Asian/Pacific Islander students is likely to be ineffective and inequitable if the former groups are differently susceptible to disability.
Effective responses to disproportionate representation of minority students in special education rely on educators demonstrating culturally competent and effective educational practices in general and special education classrooms, fully implementing the provisions of IDEA in a nonbiased manner during IEP team eligibility decisions, evaluating the out comes of minority students as served by both general and special education, and working at the community level with other agencies, parents, and business partners to respond to all the factors that affect the educational progress of minority children.
About the authors
Martha J. Coutinho, PhD, is a professor in the Department of Human Development and Learning at East Tennessee State University. Her research interests include comprehensive services for children with emotional and behavioral problems and bridging the gap between research-validated approaches, policy, current practices, and assessment.
Donald P. Oswald, PhD, is an associate professor in the Department of Psychiatry at Virginia Commonwealth University. His research interests include the system of care for students with disabilities, child behavior therapy, and developmental disabilities.
Al M. Best, PhD, is an associate professor in the Department of Biostatistics at Virginia Commonwealth University. His research interests include statistical methodology in the analysis of extant data sets.
Note: To maintain consistency, we have retained the ethnicity and disability category labels used in the OCR survey.