Predictors of minority representation in special ed

>From the web page
http://www.law.harvard.edu/civilrights/conferences/SpecEd/oswaldpaper2.h=
tml

Community and School Predictors of
Over Representation of Minority Children in Special Education


Donald P. Oswald, Ph.D.

Virginia Commonwealth University

Martha J. Coutinho, Ph.D.

East Tennessee State University

Al M. Best, Ph.D.

Virginia Commonwealth University


Abstract

The paper addresses the question of the relationship between
over representation of minority children in special education
and a set of demographic, fiscal, and school-related variables.
The intent of the study was to determine to what extent over
representation can be explained by these predictor variables.
The research was based on a conceptual framework of alternative
hypotheses regarding over representation. Hypothesis 1 proposes
that ethnic groups are differentially susceptible to disability
while Hypothesis 2 proposes that over representation is the
result of special education referral, assessment, and
eligibility processes and instruments that are culturally and
linguistically loaded and that measure and interpret the
ability, achievement, and behavior of students differently
across ethnic groups. By examining the relationship between
special education identification rates and predictor variables
we tested these hypotheses using a national sample of school
districts.

A set of logistic regression models tested the extent to which
predictor variables account for variation in disability
identification rates across ethnic groups. Variables in the full
model included gender and ethnicity of the student, and nine
district-level predictors: median value of housing, median
family income, percentage of children below poverty level,
percentage of children "at risk" in school; percentage of adults
who dropped out of school, percentage of children who are
limited English proficient, and the percentage of student
enrollment that is "non-white".  Each of the predictor variables
was found to contribute to the final model. However, the
relationship between predictor and identification rate varied
considerably across gender / ethnicity groups. For example, for
African American males and females, MR identification dropped
markedly as percent non-white increased while for White males
and females the decline is less notable. Such a differential
effect significantly influences disproportionality. Living in a
largely white community markedly increases the odds of MR
identification for African Americans, but only slightly for
White students. Further, as housing value goes from the 10th
percentile to the 90th percentile, predicted disproportionality
for Black male students increases substantially.
Correspondingly, as poverty goes from the 10th percentile to the
90th percentile, predicted disproportionality for Black male
students decreases.

           The findings demonstrate the complexity of factors
that influence special education identification.
Sociodemographic factors are clearly associated with
identification rates and with disproportionate representation
across gender and ethnic groups. Further, the effects of these
factors are often different for the various gender / ethnicity
groups and are sometimes counterintuitive. Work such as this may
serve to help identify the profiles of sociodemographic
conditions that are associated with significant disproportionate
identification. However, child gender and ethnicity also
contribute to the likelihood of identification in important
ways. This finding, along with the patterns observed in some of
the sociodemographic variables, lend indirect support to the
systemic bias hypothesis. Further study of the effects of
sociodemographic variables may contribute to exploration of bias
by highlighting the community characteristics associated with
suspect patterns of identification.

The Crisis in Minority Student Education

           The educational experience and outcomes of many
minority children in America are seriously deficient and in
special education, the problem reaches crisis proportions.  As
in regular education, the preparedness and life prospects for
many minority students with disabilities lag well behind those
of non-minority peers. In addition, there are the longstanding
concerns specific to special education: a) minority children are
disproportionately identified as disabled; b) over
representation may be the result of discriminatory or racist
attitudes and practices; and c) the benefits of special
education services are perceived as meager.

In 1994, over 1.1 million children of color were served by the
U.S. special education system (US Department of Education,
1998a). Post-high school outcomes for these minority students
with disabilities are strikingly inferior. Among secondary aged
youth with disabilities, about 75% African American students, as
compared to 47% of White students, are not employed two years
out of school.  Slightly more than half (52%) of African
Americans as compared to 39% of White young adults, are still
not employed three to five years out of school.   Within three
to five years after leaving high school, the arrest rate for
African Americans with disabilities is 40% as compared to 27%
for Whites.

In the face of such bleak outcomes, it is essential to better
understand the over representation of minority students in
special education. If this over representation is a function of
genuinely higher disability rates among students of color,
national and local responses must address the social conditions
that are risk factors for disability. If, on the other hand, the
problem arises from systemic bias and discrimination within the
public education system, aggressive efforts are required to
correct attitudes and behavior associated with special education
identification of minority children.

Our analysis of special education data suggests that both
hypotheses may be important in understanding over
representation. Statistical models of these data indicate that
social, demographic, and school-related variables are
significantly associated with special education identification.
In some cases, these relationships support the conclusion that
"toxic" social conditions may be producing disproportionately
higher rates of disability among children of color. Other
findings indicate that a significant portion of the over
representation problem may be a function of inappropriate
interpretation of ethnic and cultural differences as
disabilities.

Current Policy and Practices: An Ineffective Response

           Advocacy groups, the research community, and policy
makers have investigated, debated, and litigated the problems of
equity and over representation of minority students in special
education for over thirty years (Larry P. V. Riles, 1979, 1984,
1986; Marshall et al., v. Georgia, 1984, 1985).  There is
widespread agreement that school have failed to implement
effective responses to disproportionate representation, that is,
responses that lead to better educational experiences and
acceptable outcomes (Harry & Anderson, 1995).

The U.S. Office for Civil Rights monitors and provides
enforcement of U.S. statutes barring discrimination against
minority students in education. However, for a number of
important reasons, this strategy has been insufficient and
ineffective: a) policy responses to over representation of
minority students in a particular disability category (e.g.,
mental retardation) may lead to reduced disproportionality in
that category, but increased disproportional representation in
another category (Gottlieb et al., 1994; Oswald & Coutinho, in
press); b) keeping minority students who are already performing
poorly in the general education systems that failed them (or
inappropriately returning them there from special education)
perpetuates inferior educational outcomes for minority students
(Macmillan & Balow, 1991); c) accurate estimates of
disproportionate representation rarely inform policy responses,
and many of the available estimates too often rely on
nonrepresentative samples or faulty definitions of
disproportionality (Coutinho & Oswald, 1998; Coutinho & Oswald,
2000); and d) educators and policy makers lack sound,
empirically based information about the influence of community,
fiscal, and school-related factors on minority disability
identification rates.  Monitoring approaches do not take into
account the likelihood that demographic, school related, fiscal,
and community factors influence identification rates and that
minority children are disproportionately exposed to the
potentially toxic effects of such factors.

Improving the Special Education Process and Outcomes for
Minority Students

A critical gap exists between what is now known and what is
needed to improve the experience of minority students.  Sound,
conceptually based empirical research is essential to provide
policy makers and educators with information that can lead to
significantly improved results. Such research must: a) consider
alternative hypotheses regarding over representation in order to
improve our understanding of how community, school related and
fiscal factors influence special education identification, and
b) systematically investigate the options available for
improving the minority student experience.   Considerable
attention has been given to the hypothesis that
disproportionality is the result of biased special education
referral, assessment, and eligibility processes. Substantial
research has also been devoted to questions of instruments that
are culturally and linguistically loaded, which measure and
interpret the ability, achievement, and behavior of students
differently across ethnic groups (Gottlieb et al., 1991; Harry &
Anderson, 1995).

An alternative hypothesis is that ethnic groups are
differentially susceptible to educational disability; that is,
that the underlying distribution of educational disability
varies across ethnic groups, which, in turn, influences rates of
referral and identification as disabled.  Environmental,
demographic, health, economic, community and educational factors
may differentially affect the susceptibility of different ethnic
groups to educational disability (Coutinho & Oswald, 1998;
Coutinho   & Oswald, 2000).   A preliminary study by Oswald,
Coutinho, Best, & Singh (1998) found that a set of community and
school related variables accounted for a significant proportion
of the variability in the rate of identification of mental
retardation and emotional disturbance in African American
students, as compared to non African American students.

In sum, there is evidence to support both hypotheses, and each
is likely to influence disproportionate minority representation.
Technically sound analyses at the community level are needed to
indicate how ethnicity influenced identification for special
education, once the effects of other relevant community
variables are accounted for. Such research is required in order
to guide policy changes that assure that (a) only children who
are disabled are identified, and (b) proactive interventions
occur at the community level to achieve equity and improved
outcomes for students of color.

Purpose of the Paper

           This paper seeks to accomplish three purposes.
First, we present results of a conceptually based, empirical
study of how a set of demographic, fiscal, and school-related
factors are associated with the disproportionate representation
of minority children in special education in America. Second, we
provide specific recommendations for additional research needed
to better understand how demographic, fiscal, and school-related
variables influence disproportionality at the community level.
Finally, we offer policy implications regarding how communities
might respond to disproportionality within the context of
community characteristics (e.g. demographic profile) and school
resources.

Method

Data Sources

Every two years, the U.S. Department of Education Office for
Civil Rights collects information on a nationally representative
sample of school districts. The data are used to compile an
Elementary and Secondary School Civil Rights Compliance Report,
the chief source of data on the status of civil rights in the
nation's schools. A stratified random sampling scheme is used so
that state and national figures may be projected from the survey
data (U.S. Department of Education, 1998b). The final data set
was comprised of 4,151 school districts serving over 24 million
students in the 50 states and the District of Columbia. For this
report, we considered only the information on enrollment and
disability categories from the school year 1994-95, the most
recent survey data available at the time the study was
conducted.

The National Center for Educational Statistics, Common Core of
Data CD-ROM (NCES CCD93 Disc) has information on 15,041 school
districts in the 50 states and DC. The information in this data
set was matched with the OCR data so that only those districts
that participated in the 1994 OCR survey were included. Nine
socio-demographic variables from the Common Core of Data were
chosen as predictor variables. Table 1 provides a brief
description of each of the predictor variables, along with mean,
minimum, maximum, etc., values for the sample of districts.
Thus, for example, the median student-teacher ration for the
sample of districts was 18 students per teacher and the range of
per pupil expenditure was from a minimum of $2,263 to a maximum
of $31,625.

-----------------------------------

Insert Table 1 about here

-----------------------------------

The variables selected for these models were chosen on the basis
of several criteria: (a) the variable had been examined in
earlier work in the literature and possessed demonstrable
conceptual links to disability identification; (b) the variable
operationalized a construct about which specific predictions
could be generated, based on the alternative hypotheses being
tested; (c) the variable was included in the NCES-CCD data set;
and (d) the variable had few missing values in the NCES data
set. In addition, the variables included some community
characteristics that may be altered through political
intervention (e.g., per pupil expenditure, student-teacher
ratio) and some that are relatively "fixed" (e.g., percent
nonwhite).

Analysis Methods

We examined the effects of gender, ethnicity, and
socio-demographic factors on the proportion of students in a
school district that is identified with  Mental Retardation
(MR), Serious Emotional Disturbance (SED), or Learning
Disability (LD). The fourth disability category in this study,
"None," included students with lower incidence disability
conditions as well as all regular education students. The models
use the proportion of students in the disability category as the
dependent variable and two sets of variables as predictors: a)
the district-level socio-demographic continuous variables (see
Table 1), and b) the child-level categorical variables of gender
(F, M) and ethnicity(American Indian [AI], Asian / Pacific
Islander [AS], Black [BL], Hispanic [HI],   and White [WH]) [1].
The predictor variables were combined in a model that included
main effects and interaction effects. [2]

The analyses created logistic regression models [3] that seek to
answer the question: "Are these district-level and child-level
variables significantly associated with the likelihood of being
identified as a child with MR, SED, or LD?" More specifically,
the questions might be stated: "Does the level of poverty in the
community significantly affect the chances that a student will
be identified as mentally retarded?" or "Does being an African
American male significantly affect the chances that a student
will be identified as SED?" The multivariate model also allows
one to ask: "Taking into account the effects of poverty,
housing, per pupil expenditure, etc., do ethnicity and gender
still significantly affect the likelihood of being identified
for special education."

Results

A simple chi-square analysis of the identification data showed a
clear association between ethnicity and gender and the disabling
conditions (chi-square =3D 628,912, df =3D 27, p < .0001) verifying
the well-known fact that disability identification rates are not
the same for all of the ten gender / ethnicity groups. That is,
without taking into account the effects of social, demographic,
and school-related factors, gender and ethnicity are
significantly associated with the risk of being identified for
special education. 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 students in each of the gender /
ethnicity groups being identified as MR, SED, or 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 3.8 times as likely as White females to
be identified as SED while Black males are 5.5 times as likely
(see Table 2). These data starkly represent the extent of the
problem of disproportionality across gender and ethnic groups.

-----------------------------------

Insert Table 2 about here

-----------------------------------

A logistic regression analysis with special education
identification as the response variable and only the nine
socio-demographic variables as predictors (including linear,
quadratic, and interaction effects) was also significant
(chi-square =3D 345,130, df =3D 162, p < .0001). Thus, without
taking into account students' gender and ethnicity, the
socio-demographic conditions of a school district are strongly
associated with the proportion of students identified; some
statistically significant portion of the variation in districts'
identification rates can be explained by this combination of
predictor variables.

Given that both student characteristics (gender and ethnicity)
and socio-demographic variables were each separately associated
with the special education identification of students, we next
sought to determine whether both remained significantly
associated with identification in a combined model and whether
the relationships between the predictor variables and
identification rates were the same for each gender / ethnicity
group. A logistic regression analysis including the nine
socio-demographic variables, gender, race, and all possible
interactions was found to be significantly better than the model
with only the socio-demographic predictors (chi-square =3D
667,570, df =3D 1485, p < .0001) and significantly better than the
model with only the gender and ethnicity groups (chi-square =3D
383,788, df =3D 1620, p < .0001). There was also a significant
gender/ethnicity by socio-demographic interaction (chi-square =3D
86,224, df =3D 1458, p < .0001). These findings indicate that,
even after accounting for the effects of district
socio-demographic characteristics, students' gender and
ethnicity are important in determining the likelihood of
identification. In addition, the model demonstrates that the
impact of socio-demographic factors is different for each of the
various gender / ethnicity groups.

Predictor variables and identification rates

To illustrate the implications of the findings with respect to
public policy and best practice, we examine in greater detail
the relationship between three of the sociodemographic variables
and identification rates. For the purpose of the illustration,
we selected POVERTY, NONWHITE, and per pupil expenditure (PPE)
because they hold implicit interest with respect to the
alternative hypotheses regarding disproportionate
representation, i.e., differential susceptibility versus
systemic bias.

Poverty. The general consensus among advocates and researchers
is that increased poverty is associated with increased risk for
disability. Thus, if ethnic groups are differentially
susceptible to disability, we would expect that susceptibility
to be positively related to poverty. Ethnic groups that
experience more poverty should display increased risk for
disability and communities with more poverty should have higher
rates of special education identification. Further, across the
distribution of poverty, disproportionality may be driven in
part by the fact that children belonging to minority ethnic
groups are more likely to be found living in poverty than White
children.

On the basis of the OCR data, the logistic model estimates
identification rates for each gender/ethnicity group at every
possible value of the sociodemographic predictors. For example,
Figure 1 shows the predicted values for MR identification across
the full range of POVERTY, while holding each of the other
predictor variables at the median. Thus, the figure illustrates
the relationship between MR identification and POVERTY, when the
effects of all other predictor variables are statistically
removed (i.e., held constant).

-----------------------------------

Insert Figure 1 about here

-----------------------------------

The data reveal some unexpected findings. For example, predicted
values for MR identification among Black students decline
substantially as POVERTY increases. Further, among the
communities with the lowest POVERTY rates, the identification
rate for Black males is substantially higher than even the most
liberal prevalence estimates. Finally, disproportionality among
American Indian students, and even more strikingly among Black
students, is most pronounced in the relatively low POVERTY
communities.

For SED and LD, the relationship between identification rate and
POVERTY is in the expected direction for Black and Hispanic
students, that is, as poverty increases, identification
increases. For White and American Indian students, the trends
for LD identification are less pronounced but are generally in
the opposite direction; communities with more POVERTY tend to
identify somewhat fewer students as LD.

In sum, the POVERTY data can be viewed as supporting
differential susceptibility for SED and LD among Black and
Hispanic students. The MR data, however, appear to support an
hypothesis of systemic bias. Some portion of the
disproportionality in low POVERTY communities may be due to
White students with MR being given another disability
classification, thus artificially depressing the White rate.
Nonetheless, the absolute levels of MR identification among
Black students (especially Black males) in low POVERTY
communities suggest that a substantial number are being labeled
MR inappropriately. The situation in high POVERTY communities is
more difficult to interpret. However, the data suggest that, in
these communities, the system may be breaking down entirely such
that many students with MR go unidentified, or are given another
disability classification.

Percent of enrolled children who are non White. With respect to
NONWHITE, the expected relationship with disability
identification is null. There is no apparent rational reason to
hypothesize that living in a community that includes greater (or
smaller) numbers of ethnic minorities should represent a risk
factor for disability for students of any ethnicity. If such a
relationship is observed, and particularly if the relationship
is different for the various gender/ethnicity groups, one has
little choice but to suspect systemic bias or discrimination.

For the disability conditions MR and SED, White student
identification rates are generally consistent with rational
expectations; living in communities with greater or smaller
numbers of ethnic minorities appears to have relatively little
effect on identification. As communities become increasingly
NONWHITE, however, White students are substantially less likely
to be identified as LD. For Black students, particularly Black
male students, living in a community with few NONWHITE students
is a substantial risk factor for MR and SED identification,
leading to marked disproportionality at that end of the
distribution. Conversely, American Indian students living in
high NONWHITE communities have substantially higher
identification rates, particularly for SED. Figure 2 illustrates
the relationship between NONWHITE and SED identification.

-----------------------------------

Insert Figure 2 about here

-----------------------------------

In sum, the expected null relationship between disability
identification and NONWHITE was not observed. The findings
indicate a need to carefully scrutinize SED and MR
identification in low NONWHITE communities with an eye toward
detecting inappropriate identification that is based more on
difference than on disability. High NONWHITE communities require
some careful consideration with respect to SED identification
among American Indian students.

Per Pupil Expenditure. Per pupil expenditure might be expected
to have some relationship to identification in that schools that
spend more money per pupil should be more likely to identify
appropriate numbers of students with disabilities and less
likely to sustain special education eligibility processes that
are systematically biased.

The observed relationships between identification rates and PPE,
however, are complex. For students with SED, the trends match
expectations reasonably well; systems that spend more also
identify more students with SED. Disproportionality does not
vary dramatically across the distribution except for Hispanic
and American Indian males, where increased PPE substantially
increases disproportionality for these two groups.
Disproportionality among students with MR also tends to increase
across the distribution for Black females and Hispanic students.
At the low end of the PPE distribution (<$5,000),
disproportionality appears to be exacerbated for Black and
Hispanic students with LD.

In sum, PPE appears to have a modest effect on identification.
Communities that spend less money have somewhat more
disproportionate identification of Hispanic and African American
students as LD. However, for some gender / ethnicity groups
(e.g., Black females and Hispanic students, with respect to MR),
communities that spend more for education show greater
disproportionality, making it clear that increasing school
expenditures does not necessarily result in improvement in
disproportional representation. While increased overall
education expenditure may result from identifying more children
for special education, there is no clear rational explanation as
to why it should be associated with increased disproportionality.

Implications and Recommendations

           The findings reported above demonstrate the
complexity of factors that influence special education
identification. Sociodemographic factors are clearly associated
with identification rates and with disproportionate
representation across gender and ethnic groups. Further, the
effects of these factors are often different for the various
gender / ethnicity groups and are sometimes counterintuitive.
However, work such as this may serve to help identify the
profiles of sociodemographic conditions that are associated with
significant disproportionate identification.

           In spite of the importance of sociodemographic
factors, however, child gender and ethnicity also contribute to
the likelihood of identification in important ways. This
finding, along with the patterns observed in some of the
sociodemographic variables, lend indirect support to the
systemic bias hypothesis. Further study of the effects of
sociodemographic variables may contribute to exploration of bias
by highlighting the community characteristics associated with
suspect patterns of identification.

Policy makers and educators need access to information that
provides a profile of how community and school resources and
other sociodemographic factors may contribute to the
disproportionate representation of minority students in special
education.   Toward that end, a conceptually and empirically
guided research agenda is needed to disentangle effects related
to differential susceptibility from those related to systemic
biases in the special and regular education systems.  Such
research is needed at the community level to provide knowledge
regarding the significance of disproportionality and
recommendations regarding how to reform educational practices in
a manner that yields equitable and effective educational
experiences and improved education outcomes for all students.

References

Coutinho, M.J., & Oswald, D. P.  (1999).   Ethnicity and special
education research:  Identifying questions and methods.
Behavioral Disorders, 24, 66-73.

Coutinho, M. & Oswald, D.   (2000).   Disproportionate
Representation in Special Education: A Synthesis and
Recommendations.   Journal of Child and Family Studies, 9,
135-156.

Draper, N.R., & Smith, H. (1998). Applied regression analyses.
New York: John Wiley & Sons

.

Gottlieb, J., Gottlieb, B.W., & Trongue, S.  (1991)   Parent and
teacher referrals for a psychoeducational evaluation. The
Journal of Special Education, 25, 155-167.

Gottlieb, J., & Alter, M. (1994).  An analysis of referrals,
placement, and progress of children with disabilities who attend
New York City public schools.  New York University, NY:  School
of Education.  (ERIC Document Reproduction Service No. ED 414
372).

Harry, B., & Anderson, M. G.  The disproportionate placement of
African American males in special education programs:   A
critique of the process.   Journal of Negro Education, 63,
602-619.

Hosmer, D. W., & Lemeshow, S. (1989). Applied logistic
regression. New York: John Wiley & Sons.

Larry P. v. Wilson Riles.   343 F. Supp. 1306 (N.D. Cal. 1972)
(preliminary injunction).  Aff'd 502 F. 2d 963 (9th cir. 1974);
495F. Supp. 926 (N.D. Cal. 1979) (decision on merits).  Aff'd
(9th cir. No. 80-427 Jan. 23, 2984). Order modifying judgment,
C-71-2270 RFP, Sept. 25, 1986.

Macmillan, D. L.,   & Balow, I. H.   (1991). Impact of Larry P.
on eudcaitonal programs and assessment practices in California.
 Diagnostique, 17, 57-69.

Marshall et al., v. Georgia.   U.S. District Court for the
Southern district of Georgia, CV482-233, June 28, 1984.

NAEP.   National Assessment of Educational Progress.  (199).
Accelerating academic achievmeent; A summary of findings from 20
years of NAEP.   Wasington, D.C.:   U.S. Department of Education.

Oswald, D., Coutinho, M., Singh, N., & Best, A. (1998).
Ethnicity in special education and relationships with school
related economic and educational variables. Journal of Special
Education, 32, 194-206.

Oswald, D.P., & Coutinho, M.J.  (in press).   Trends in
disproportionate representation in special education:
Implications for multicultural education policies.  In C.A.
Utley & F.E. Obiakor (Eds.), Special education, multicultural
education, and school reform: Components of a quality education
for students with mild disabilities.  Springfield, IL: Charles
C. Thomas.

SCANS (1991).   What work requires of schools.   A SCANS report
for America 2000.  Washington, DC: Secretary's Commission of
Achieving Necessary Skills (SCANS), U.S. Department of Labor.

US Department of Education. (1998a). Twentieth annual report to
Congress. Washington, D. C.: Author.

           US Department of Education. (1998b). Office for Civil
Rights - Fiscal Year 1998: Annual Report to Congress.
Washington, DC: Author.

Authors Note:

Preparation of this manuscript was supported in part by the
Field-Initiated Studies Program of the National Institute on
Educational Governance, Finance, Policymaking, and Management,
Office of Educational Research and Improvement, U.S. Department
of Education (Grant No. R308FG70020).

Table 1.

Summary of socio-demographic predictors

N

Predictor

Predictor Description

Mean

S.D.

Minimum

10%tile

Median

90th%tile

Maximum

STR

Student-teacher ratio (# of enrolled students divided by # of
teachers employed)

18.527

5.107

0.4

15.3

18

24.6

108

PPE

Per pupil expenditure (expenditure for regular and special
education divided by # of enrolled students)

5059.807

2009.141

2263

3663

4756

7313

31625

ATRISK

Percent of children enrolled who are "at risk"

4.013

5.354

0

0.5

3.6

10.1

63.2

NONWHITE

Percent of enrolled children who are non White

24.10

30.342

0

3

25.3

61.3

100

LEP

Percentage of enrolled children who are Limited English
Proficient

2.109

4.221

0

0.2

1

6.4

35.6

HOUSING

Median housing value for houses, in $10,000 units

9.503

8.864

0.75

4.2733

7.55

19.2027

48.3776

INCOME

Median income for households with children, in $100,000 units

0.363

0.164

0

0.2441

0.325

0.50115

1.40555

POVERTY

Percent of children in households below poverty level

17.832

15.545

0

5.5

18.2

33.6

100

NO DIPLOMA

Percent of adults in the community who have education of 12th
grade or less and no diploma

25.294

15.489

0

10.9

24.5

39.2

100

Table 2

Identification Odds Ratios for Gender / Ethnicity Groups

Gender

Ethnicity

MR Odds Ratio

LD Odds Ratio

SED Odds Ratio

M

American Indian

1.66

2.903

5.024

M

Asian / Pacific Islander

.50

0.783

0.915

M

Black

3.26

2.343

5.527

M

Hispanic

.95

2.104

2.354

M

White

1.36

2.279

3.810

F

American Indian

1.21

1.339

1.374

F

Asian / Pacific Islander

.40

0.350

0.229

F

Black

2.02

0.978

1.376

F

Hispanic

.70

1.021

0.588

F

White

1.00

1.000

1.000

Figure Captions

Figure 1. MR Identification rates and poverty

Figure 2. SED Identification rates and percent nonwhite


[1] To maintain consistency, we have retained the ethnicity and
disability category labels used in the OCR survey.

[2] The continuous predictor, or covariate, 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 '
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, it was decided to use p < .0005 as the cut-off for
significance.

[3] Categorical responses (such as Disability Condition) may be
modeled using logistic regression (Hosmer & Lemeshow, 1989) and
such models include the proportion in each disability category
as the response variable. For all analyses, districts were
weighted by the number of students (as well as by sample weight)
so that the models simulate using the student as the unit of
analysis rather than the district.

----------
End of Document





-- 
TNET Mail-To-News Gateway Version - 1.6
For information about this gateway email programs@tnet.com