Beruflich Dokumente
Kultur Dokumente
Professor Reimers
Applied Econometrics
Final Paper
9 December 2016
INTRODUCTION
In 1954, the issue of education inequality was brought to the forefront of the American
consciousness with the Supreme Court ruling Brown v. Board of Ed. The landmark ruling
declared separation inherently unequal and forbade the segregation of schools based upon race.
Overruling Plessy v. Ferguson, the Brown ruling found that it was impossible to forcibly
segregate people by race and still provide equal treatment. While the Brown decision was huge,
it focused very specifically on the legal aspect of segregation, alleging that it was the legal
exclusion of blacks that created the inequality, not necessarily the disparity in conditions that
followed legal racial segregation. The issue with this view of the subject is that legal segregation
in this country did not fall under the umbrella of separate but equal as stipulated in Plessy v.
Ferguson, but in fact a large number of inequalities came about as a result of the combination of
forced segregation and lack economic opportunities for black Americans, in addition to the
neglect of public facilities designated for blacks. The end of formal school segregation policies
was by no means an end to school segregation, nor an end to unequal conditions in schools for
different races. As a result, there exist significant disparities in academic performance between
Because education is such an integral part of ones human capital stock, eliminating
disparities in academic performance between black and white students may be one of the most
important steps toward eliminating many other racial disparities that exist in this country. While
many studies have been done analyzing the black-white achievement gap and its causes, the gap
has remained significant since it was first measured in 1965. The goal of this paper is to analyze
the black-white achievement gap in Massachusetts, using data from the 2016 PARCC exam in
addition to other school data from the Massachusetts Department of Educations statewide
reports.
LITERATURE REVIEW
The first major inquiry into race-based inequities in education is known as the Coleman
Report, and was conducted in 1965. By conducting a survey of public schools around the nation,
Coleman found that black students were systematically underperforming compared to white
students, in addition to finding evidence of extreme racial segregation in the school system. He
found that minority scores were as much as 1 standard deviation lower than those of white
children. He also discovered 65% of black children grades 1-12 attended a school that was 90-
100% black, as well as 80% of white students attended a school that was 90-100% white. These
numbers tended to increase with age, and in the South many students attended schools that were
completely racially homogenous. In addition, he found that black children were more likely to
have larger class sizes, attend schools with less robust facilities, have unenthusiastic teachers,
and be missing useful educational resources, like an encyclopedia in the home. Based on his
findings, Coleman attributed much of the cause of the gap to the effects of segregation, and
inequities in school quality between black and white children (Coleman, 1965).
Since Colemans study, the gap has remained but the size of the gap has fluctuated,
experiencing a decline throughout the 1970s and 80s, but becoming stagnant in the 90s and
possibly starting to increase again. There have been a variety of different studies aimed at
identifying and understanding the causes, to varying degrees of success. In 2004, using the
ECLS-K data, Fryer and Levitt found no gap at the kindergarten level when controlling for a
small amount of variables related to socioeconomic status. They found that the gap returned by
3rd grade, however, and continued to get worse as school progressed. As with Coleman, they
found that the gap seems to consistently grow as students age. They explored different reasons
for the increase in the gaps over time, and found that large amounts of segregation and inequality
between schools attended by whites and blacks were still present, however they could not find
any conclusive evidence as to what exactly was driving the gap in achievement to widen over
time, given that even when controlling for these factors black children still underperform as
determine the ways in which segregation influences the academic achievement gap, particularly
looking at whether it was the racial or socioeconomic composition of schools that tended to
affect minority children. He analyzed standardized testing data from the National Center for
Education Statistics (NCES), along with data on school segregation from the Common Core of
Data (CCD), and data on residential segregation from the American Community Survey (ACS).
Ultimately he found the disparity in average poverty rates between schools attended by blacks
and whites to have the strongest correlation to the academic achievement gap (Reardon, 2015).
The goal of this paper is to use some of the previous techniques for analysis, and apply
them to data from schools throughout Massachusetts in order to get an idea of the size of the
achievement gap in Massachusetts, as well as identify some of the factors most strongly
DESCRIPTION OF DATA
The data used for this study comes from the Massachusetts Department of Educations
yearly statewide reports. They provide data on a number of school, teacher, and student related
variables, broken down at either the school or district level. For this paper, data on PARCC
achievement results were used as measures of students academic performance. The PARCC
(Partnership for Assessment of Readiness for College and Careers) is a yearly standardized test
for English/Language Arts and Mathematics that is designed to accommodate for more rigorous
education standards that have been adopted by many states in recent years. The Massachusetts
statewide reports provide the average scaled scores for students in each racial category, and these
are used as the dependent variable in the model. The lowest possible score is a 650, while the
highest is an 850, and the score is a measure of how prepared a student is for the next grade
level, an 850 representing a student who well exceeds expectations, while a 650 represents a
student who is not prepared for the next grade. In addition to the average scaled PARCC scores,
school level data on several different school and student related variables was used for the
independent variables. The percentage of full-time equivalent (FTE) black, Asian, and Hispanic
teachers, percentage of teachers at each evaluation level, and average class size at each school
were used as measures of school quality. The percentage of black, white, and economically
disadvantaged students in each class was used to see how the racial and socioeconomic balance
Below are the descriptive statistics for the data used. Not all schools reported average
scaled scores for both black and white students, so two tables are shown, the first with the
average scaled scores for black students and school data from the schools that reported them, and
the second with the averaged scaled scores for white students and the school data for the schools
that reported them. Ill refer to the first set of data as the black student dataset and the second as
Avg
Average Class
Statistic Scaled Score size % Black %White %Disadvantaged
Nbr. of
observations 191 191 191 191 191
Nbr. of missing
values 0 0 0 0 0
Minimum 709.000 10.300 0.000 8.800 3.500
Maximum 796.000 25.400 73.500 97.400 77.200
Range 87.000 15.100 73.500 88.600 73.700
Median 753.000 18.700 3.800 71.200 24.900
Mean 752.848 18.626 8.377 64.705 28.841
Standard
deviation (n-1) 16.396 2.891 12.042 25.761 18.486
% Black
Teachers % Hispanic % Asian % % Needs %
Teachers Teachers Exemplary % Proficient Improvement Unsatisfactory
191 191 191 191 191 191 191
0 0 0 0 0 0 0
0.000 0.000 0.000 0.000 40.000 0.000 0.000
22.388 13.492 8.451 52.300 99.500 36.100 13.600
22.388 13.492 8.451 52.300 59.500 36.100 13.600
0.000 0.000 0.000 7.900 87.500 3.200 0.000
1.350 1.773 1.204 9.951 84.702 4.666 0.683
2.963 2.782 1.873 9.282 10.493 5.140 1.661
As shown in the tables above, the means of the Average Scaled Scores for black and
white students are 730.425 and 752.848 respectively, meaning that the overall difference in
Average Scaled Score between black and white students in these samples is 22.423. It is worth
noting that the white mean Average Scaled Score of 752.848 is just past the level 4 cutoff of 750,
while the black mean Average Scaled Score is solidly in the level 3 range. According to the
PARCCs scoring guide, a level 4 student is meeting expectations, while a level 3 student is
approaching expectations. The white students scores also had slightly less variation than the
black students. The mean average class size was similar for each of the datasets, but for the
black student data set the range and standard deviation were larger showing that there was more
inconsistency in the average class sizes for black students. The two datasets were very different
when it came to the percentages of black, white, and economically disadvantaged students. The
white student dataset shows had a mean black percentage of 8.337, mean white percentage of
64.705, and mean economically disadvantaged percentage of 28.441, compared to 25.26, 34.86,
and 41.66 respectively for the black student dataset. This shows that white students attend more
racially homogenous schools in general, and on average are around wealthier classmates than
black students. The mean percentages of FTE black, Asian, and Hispanic teachers were very low
for both datasets, indicating that in general there is a severe lack of diversity among teachers, and
that white teachers are extremely overrepresented throughout the state. Despite this, the black
student dataset had higher means for all minority teacher percentages, particularly black teachers.
The percentages of teachers at each evaluation level was about the same for each dataset.
MODEL
The dependent variables for this model will be the Average Scaled Scores for both black
and white students. In order to determine what factors might affect the black-white score gap I
will analyze the effect of each of the independent variables on the Average Scaled Score for both
black and white students, and compare the different coefficients to determine which areas might
best be targeted by policy makers. The independent variables include the percentage of FTE
black, Asian, and Hispanic teachers, the percentage of teachers at each evaluation level, the
percentage of black, white, and economically disadvantaged students, and the average class size
Based upon economic theory, and the relevant literature, I would predict that for the
black student dataset, the percentages of FTE minority teachers, particularly black teachers,
would have positive coefficients. Because teachers represent positive role models, having more
identifiable role models in the classroom might improve performance. For white students, while
the positive minority role model theory would not apply to them, I also would not predict the
percentages of FTE minority teachers to have a negative impact, therefore Im predicting that for
the white student dataset these variables will be insignificant. For the percentages of teachers at
each evaluation level, coefficients should be positive for both exemplary and proficient
categories, and negative for needs improvement and unsatisfactory categories. These should
be the same for both the white and black student datasets. For the black student dataset, I predict
that the % of Black students in the classroom should have a positive coefficient, as students
would feel more comfortable when around more students of the same racial group, that the % of
white and economically disadvantaged students should have negative coefficients. For the white
student dataset, I would predict that the % of white students in the classroom should have a
positive coefficient, while the % of black and economically disadvantaged students would both
have negative coefficients. Average class size should have a negative coefficient for both
datasets.
Estimating The Model
R2 0.255 0.425
R2 0.147 0.306
first set of regressions, coefficients for all of the independent variables were calculated using
OLS regression for the Average Scaled Scores of each dataset. As shown, the model was only
able to explain about 12.5% of the variation in Average Scaled Score for the black student
dataset, compared to 39% for the white student dataset. The white student dataset contained more
data due to the fact that there were more schools reporting White Average Scaled Scores, which
is likely why the results were better. There were only two coefficients that were significant for
the black student dataset, the percentages of white and economically disadvantaged students, -
0.457 and -0.613 respectively, but they were both highly significant. They also are both negative
as hypothesized. These coefficients mean that for a 1% increase in the percentage of white or
economically disadvantaged kids in the average class, the Average Scaled Score for black
students decreases by 0.457 points and 0.613 points respectively, holding all else equal.
Considering that the mean % white is 34.86 and the mean % economically disadvantaged is
41.67 for schools reporting Black Average Scaled Scores, it seems the average black student is
heavily affected by these factors. For the white student dataset, there were more significant
coefficients, in addition to the % white and % economically disadvantaged which were still
highly significant. The % Black students was significant at the 1% level, and had a negative
coefficient as expected, but the % white also had a negative coefficient, which was surprising.
The % economically disadvantaged students had a negative coefficient as expected. The percent
of FTE Black teachers was also surprisingly significant at the 0.5% level, and even more
confusingly had a positive coefficient of 0.852, meaning that a 1% increase in the percentage of
FTE Black teachers would actually lead to an increase in the Average Scaled Scores of white
students by nearly a whole point holding all else equal. This relationship was not expected, and is
hard to interpret as none of the relevant literature or theory seems to suggest that the amount of
Black teachers a white student is exposed to would improve their academic performance, but
Because the variables for the percentages of teachers at each evaluation level were all
insignificant, I decided to omit them from the second model, as they were likely redundant. In
order to test for heteroskedasticity, the Breusch-Pagan test was done. For both datasets the P-
value obtained was larger than the critical alpha of 0.05, meaning that we fail to reject the null
hypothesis of homoskedastic residuals. In addition to the Breusch-Pagan test, I looked at the VIF
for each independent variable to determine if there was significant multicollinearity present. For
each dataset, the % White students had a very high VIF, and was very highly correlated with the
White students resulted in normal VIFs for all variables, so it was also omitted from the final
model.
After running OLS regressions with the discussed changes to the model, the R2 and the
adjusted R2 fell for both equations, meaning that overall it explained less variation in students
Average Scaled Scores. The % of Economically Disadvantaged students in the classroom was
still significant at the 0.05% level for both datasets, however for the black student dataset it is
now the only significant variable. The % of FTE Hispanic teachers is now significant at the 5%
level and is positive. Notably, while the coefficient for % Economically Disadvantaged is
negative for both the black and white student dataset, it is almost twice as large for the white
students in the classroom causes the Average Scaled Scores of white students to decrease nearly
twice as much as black students. The positive coefficient for the % of FTE Hispanic teachers is
hard to interpret, similar to the previously positive coefficient for % of FTE Black teachers,
Because not many of the variables were not significantly correlated with the Average
Scaled Scores for either black or white students, I would prefer to attempt another study using a
larger dataset. Provided more time and access, using the NAEP restricted dataset would allow for
a much larger sample of students from high schools across the country, as well as a richer set of
data describing those students. The explanatory variables used in this model were to a certain
extent used because they were the only available pieces of data, in addition to the fact that my
review of the related literature led me to believe that they might have an effect on the black-
white score gap. In addition to the fact that higher quality data could improve the study, I also
believe that using PARCC data could have had a negative impact on the results. I was unaware
when I began my study, but it seems that students have the option to opt-out of the test, thus as a
result my data may not be as useful as I had thought because the Average Scaled Scores are
based on smaller samples of students from each school. Because of the source of my data, and
the fact that the only information I saw about opting out was on websites operated by parent
advocate groups, it was hard to determine the degree to which this affected the results reported.
However, it seems as though the participation in the MCAS test is generally higher, so it likely
would have been better to use MCAS scores. This is an additional reason why a more detailed
dataset would have made for a much more effective study, though participation and quality
Despite the issues and poor fit of the models, some of the results could help to steer
future studies. In particular, the fact that white students were impacted much more by the
result. While the data shows that black students are much more likely to be in classrooms with a
performance. Including dummy variables for the type of school students attend might provide
more insight, as it is possible that wealthier white students are able to economically segregate
themselves more by attending private or charter schools while wealthier black students are not,
thus resulting in a greater gap between wealthy white students and economically disadvantaged
students in general. The unexpectedly positive coefficients for % of FTE Black teachers, and %
of FTE Hispanic teachers in the 1st and 2nd regressions respectively were hard to interpret since
there was no theoretical relationship from the literature. Its hard to understand why either of
these variables would have a positive impact on white students Average Scaled Scores,
especially given that the other minority teacher variables were insignificant in both cases. With
access to data related to the racial makeup of students residential areas in addition to the makeup
of their schools, its likely I would get better results, as my current model was based upon the
effect of representation in the classroom, but residential factors definitely have an effect on
students performance.
Lastly, because of the issue I discovered with students opting out of the PARCC test,
which was a movement that seemed to be driven by parent groups, I would suggest that some
effort be made to rebrand these tests so that they can be useful data collection tools that would
help improve education instead of infuriate parents. In particular, the link between school test
results and school funding seemed to be an issue for many parents because of the perception that
it forced teachers to tailor curriculums to the tests, rather than focusing on providing the highest
quality education. Changing public attitudes towards these types of data gathering tools could
help improve the quality of the data available, allowing researchers to provide useful analysis
<http://files.eric.ed.gov/fulltext/ED012275.pdf>.
Roland Fryer, and Steven Levitt. "Understanding the Black-White Test Score Gap in the First Two
<http://cseweb.ucsd.edu/~ddahlstr/misc/test-score-gap.pdf>.
Reardon, Sean F. "School Segregation and Racial Achievement Gaps." CEPA Working Paper No.
15-12 (2015): n. pag. School Segregation and Racial Achievement Gaps. Center for
<https://cepa.stanford.edu/sites/default/files/wp15-12v201510.pdf>.
"Score Results." PARCC. Partnership for Assessment of Readiness for College and Careers, n.d.
"Fact Sheet on Opting Out of MCAS/PARCC Testing." Citizens for Public Schools. Citizens for
facts-on-opting-out-of-mcas-or-parcc/>.