Professional Documents
Culture Documents
Joseph Johnson
I. Hypothesis
I will test if a larger student-teacher ratio negatively affects the
Average Freshman Graduation Rate (AFGR) in high schools.
II. Motivation
Having a high school diploma is an important asset to have. Studies
have shown that people who do not graduate high school have a
harder time finding well paying jobs, have more health issues, have a
higher chance of being in the court system, are more likely to be on a
social welfare program, and their children are more likely to not
graduate high school either, potentially leading to an endless cycle of
poverty. Being that a high school diploma has this much effect on ones
life it would be logical to look at factors that influence graduation rates.
Student-teacher ratio (STR) has been an issue that has be discussed
when talking about improving education. Many studies have been
conducted on the effects of student-teacher ratio on student
achievement. This is because states have encountered problems
balancing their budget as 40 states are projecting shortfalls for their
2012 budget year and state funding of public education could be on
the chopping block. This might make sense that education would be on
the chopping block seeing that a study found that increasing the STR
by one in every state would cut costs by $12 billion in teacher salary
costs alone (Whitehurst, 2011).
On top of this, there is an issue with local school funding. Because
public schools are funded by property tax incurred by the people who
live near the districts. This would cause school districts that have
people of higher economic status that live near that district to have
substantially more funds than districts who have people that live near
them that are worse off. This disparity in revenue via property tax was
such an issue that in Washingtons Supreme Court ruled in McCleary v.
State of Washington that this way of funding schools was
unconstitutional and the State must provide stable and dependable
funding for such costs. Also, funding must be based as closely as
reasonably practicable on actual costs. A schools budget was shown to
be important in a study by the Albert Shanker Institute. The study
revealed that spending matters in education so places that have
people of higher economic status would be more likely to have
successful students than those who have people that are worse off. All
these issues make it very rational to analyze the benefit of STR.
Literature Review:
The paper I chose was a paper by the Brookings Institute, Mr. Grover J.
Whitehurst and Matthew M. Chingos they examined several research
papers that analyzed the effect of student teacher ratio (STR) on
student achievement. In the paper they pointed out that the student-
V. Regression Model:
The five models I ran were:
Regression 1:
AFGR = 0 + 1(STR) + 2(Perblack) + 3(STR*Black )
Regression 2:
AFGR = 0 + 1(STR) + 2(Perblack) + 3(STR*Black )
+4(Black)
Regression 3:
AFGR = 0 + 1(STR) + 2(Perblack)
Regression 4:
AFGR = 0 + 1(STR) + 2(Perblack) + 3(Adjfl)+4AdjRev
Regression 5:
AFGR = 0 + 1(STR) + 2(Perblack) + 3(Adjfl)+4AdjRev + 5(Strblack) +
6(black) + 7(Exit)
Where i represents the percent change in the public high school students who
graduate with a regular diploma given a one-unit change in Xi.
VI. Results
In my model there was only three variables that were consistently significant at the 95%
level were STR, Exit and Perblack. The parameter estimate for strblack had an
unexpected value attributed to it.
Student-Teacher Ratio:
In my models the parameter estimate for Student-Teacher ratio was consistently negative
and statistically significant. There was a small difference in the magnitude of the
parameter estimates between models.
Adjusted Free Lunch:
In my models, I saw large parameter estimates because it had possible values from 0 to 1.
So, if the state only contained children that were on subsidized lunch then it would be
one. It would be hard to interpret the meaning of the parameter estimate.
Perblack:
In my models, this variable lie, STR, was consistently negative and there was little
difference in the magnitudes.
Adjusted Local Revenue:
In my models, AdjRev was significant for one of my models and not for
the other model I ran that contained the variable. The signs on the
variables parameter estimates were what I predicted.
Exit:
The parameter estimates for the variable Exit were consistently significant at the 95%
level. The signs on the variables parameter estimates were what I
predicted.
STR*Black:
The parameter estimates for this variable was only significant at the
90% level once. It was not statistically significant in any other of the
models.
VII. Conclusion
Assuming that Model (4) in my table is the best model of the models I
regressed. The study shows that student-teacher ratio to be
statistically significant in determining the graduation rates of high
school students. Also, it seems that public high school students who
graduate with a regular diploma would decrease -0.44 percent given a
one-unit increase in STR. I do not believe that going off of this model
alone a lawmaker would find it beneficial to decrease STR given how
low the magnitude of percent change in graduation rates that would
cause and given the apparent cost savings of cutting a teacher.
My model has several flaws in it. My model does not actually address
the income in that state, which would probably be an important
variable in the regression. It deals with state data instead of data that
goes by district, which would probably decrease the accuracy of the
data that I am receiving. I did not put in a variable that determined the
change in AFGR if the teacher cut was a bad teacher or a good teacher.
Variable
Description
Source
AFGR
An statewide
estimate of
the
percentage
of public
high school
students
who
graduate
with a
regular
diploma 4
http://nces.ed.gov/pubs2011/graduates/tables/table_03.a
years after
starting
AdjFl
STR
Exit
AdjRev
black
Perblack
The percent
of students
on free or
reduced
lunch
adjusted by
state
population.
Average
number of
students a
teacher has
in class
A variable
that states
whether or
not a state
has an exit
exam
Local
revenue
adjusted by
state
population in
2008
A variable
that
determined
if the state
had a higher
percentage
of black
people than
the national
average
Percentage
http://nces.ed.gov/programs/digest/d10/tables/dt10_044
http://www.census.gov/statab/ranks/rank01.html
http://nces.ed.gov/pubs2010/2010309/tables/table_04.as
http://www.edweek.org/rc/articles/2008/11/05/sow1105.h
http://www.census.gov//govs/school/historical_data_2008
http://www.census.gov/statab/ranks/rank01.html
http://quickfacts.census.gov/qfd/states/00000.html
http://nces.ed.gov/pubs2010/2010309/tables/table_02.as
black
referrer=report
public
Student-Std Dev - Minimum
-0.48** Maximum
Variable
Label
N schools
Mean
Teacher AFGR0.72**
0.75** 0.71**
0.44** (0.20)
AFGR
50 76.1900192
7.251062
56.3 89.5999756
Ratio
(0.25)
(0.20)
(0.19)
(0.175
STR
STR
50 11.9283738
3.5488933
6.754592 23.3856213
5
(STR)
)
Exit
Exit
50
0.46
0.50345741
Number of
-0
Statepo
Statepop
50
6070157.82
6749347.82
532668
36756666
students on
64.41* 66.23*
pa
*
(33.24
AdjRev
AdjRev
50
0.7580078
0.3299779
0.059507
1.586908
subsidized
(31.59
)
9
lunch PerBlack 50 14.5490594 12.9882432) 0.955728 50.6010661
PerBlack
program
4
adjusted by
adjfl
50
0.1567138
0.0148334
0.108788
0.1924197
5
State
black
50
0.44
0.5014265
0
1
Population
strblack
50
4.7626764
5.5874587
0
15.1674004
(Adjfl)
Percent of
-0.32**
black
0.44** 0.43** 0.32** 0.24** (0.11)
students
(0.10) (0.09) (0.06) (0.06)
(PerBlack)
STR*Black 0.39*
0.56
0.37
(Strblack)
(0.21) (0.48)
(0.47)
State has a
-2.27
-1.12
black
(6.43)
(6.14)
population
higher than
the median
(Black)
Exit Exam
-5.01**
(Exit)
4.95** (1.69)
(1.60)
Local
4.83**
3.71
Revenue
(2.22)
(2.4)
Adjusted
By State
Population
References:
(AdjRev)
Intercept
89.41* 89.74* 89.24* 93.74* 91.08*
*
*
*
*
*
(3.23) (2.97) (2.78) (6.29) (8.08)
Summary Statistics
SER
1611.92 1609.5 1709.8 1227.5 1454.8
4
5
8
4
N
50
50
50
50
50
0.334
0.320
0.308
0.469
0.357
(*) Indicates that the parameter estimate is
significant at the 90% level. (**) Indicates that the
parameter estimate is significant at the 95% level.
The numbers in the parenthesis are the standard
error of the parameter estimates.