You are on page 1of 7

Question 6.

Tennessee's Student Teacher Achievement Ratio (STAR) project was a largescale, four-year experimental study designed to evaluate the effect on
learning of reduced class sizes. The study compared three alternative class
size arrangements for Kindergarten through Grade 3:

regular class size, with 22 to 25 students per class, a single teacher


and no aides
small class size, with 13-17 students per class and no aide
regular class size plus a teachers aide.

Refer to the documents in folder Question 5 for more information. Read the
data description carefully as you need to understand how different variables
are constructed and what different values they take. Some of these variables
you are going to use below have missing values that you need to exclude
when you are analyzing the data.
(A) Use the dataset star and report simple differences estimates of the
impact of the two treatments relative to the base case of a regular class size
with no teachers aide for students attending project STAR classes. Use as an
outcome measure the students combined score on the math and reading
portions of the Stanford Achievement Test. Show these results for each grade
level (i.e. kindergarten, grade1, grade2, grade3) in a table, together with
standard errors. Indicate which estimates are statistically significant at the
1% and 5% significance levels. Also report the sample sizes in the table.
PartA
Small Class
Regular Class with Teacher Aide
N
R2

(1)
Kindergarten
13.90***
(2.409)
0.314
(2.310)
5786
0.007

(2)
Grade1
29.78***
(2.807)
11.96***
(2.686)
6379
0.017

(3)
Grade2
19.39***
(2.710)
3.479
(2.566)
6049
0.009

(4)
Grade3
15.59***
(2.395)
-0.291
(2.302)
5967
0.010

Standard errors in parentheses


*
p < 0.10, ** p < 0.05, *** p < 0.01

(B) Add controls for gender, race, quarter of birth, school type and socioeconomic status in regressions for each grade level, and report in a table the
same set of impact estimates as reported in part (A). Interpret your
estimated coefficients for these control variables. Are there any statistically
significant differences between the estimated treatment effects without
controls and those with controls for different grades? What do you infer from

the differences (or absence of differences) between these two sets of


estimated impacts?

Small Class
Regular Class + Aide
female
White
Black
Asian
Hispanic
1st Quarter of Birth
2nd Quarter of Birth
3rd Quarter of Birth
Inner City School
Suburban School
Urban School
Free Lunch
N
R2

(1)
Kindergarten
14.95***
(3.182)
-0.0909
(3.117)
9.024***
(2.578)
-83.85*
(46.58)
-103.1**
(46.81)
-120.0**
(52.83)
-75.01
(65.80)
-8.608**
(3.724)
-19.86***
(3.742)
-29.53***
(3.757)
-2.904
(7.499)
-8.453
(6.332)
-14.11**
(5.813)
-24.96***
(3.104)
2617
0.095

(2)
Grade1
22.72***
(3.565)
6.589*
(3.674)
6.198**
(2.938)
-66.51
(53.84)
-89.35*
(54.10)
-44.66
(61.09)
24.03
(76.07)
-9.143**
(4.262)
-17.28***
(4.286)
-26.24***
(4.277)
-39.94***
(8.546)
-17.45**
(7.263)
-7.280
(6.696)
-42.21***
(3.528)
2697
0.202

(3)
Grade2
13.09***
(3.577)
2.899
(3.652)
6.776**
(2.871)
-123.8**
(52.66)
-158.0***
(52.90)
-82.04
(59.73)
-105.2
(74.40)
-5.251
(4.158)
-10.09**
(4.181)
-18.43***
(4.167)
-18.05**
(8.356)
-10.29
(7.102)
1.698
(6.558)
-34.13***
(3.435)
2696
0.171

(4)
Grade3
8.753***
(3.324)
-0.342
(3.412)
6.465**
(2.606)
-63.07
(47.17)
-89.56*
(47.40)
-50.61
(53.50)
-13.32
(66.64)
-6.893*
(3.777)
-7.481**
(3.791)
-13.21***
(3.803)
-9.663
(7.576)
-5.377
(6.416)
-4.480
(5.906)
-33.14***
(3.158)
2629
0.136

Regarding the effect of small class size, the estimated coefficients for
Kindergarten students dont change much, but for all other grades they
decrease significantly across the board when we add the controls to the
model. This suggests that part of the better performance of students in
smaller classes has to do with the observed characteristics that are

correlated with class size. In other words, students in smaller classes tend to
have better characteristics that explain part of their better performance.
Regarding the effect of regular class size with teacher aide, the only
coefficient that changes significantly is the one for grade1, which suggests
the same thing argued above.

(C) Write down the regression equation you would run to estimate whether
the effect of the two treatments is different for black versus white students
(Hint: keep in mind that there are races other than black and white in the
data. We want those observations to be included in the regressions, but we
only want to compare the blacks and whites. One way to do this is to lump all
other races into one category (using an indicator) and including them as one
group
(e.g.
other
races)
in
your
regression).
For each grade level, run this regression (use the same control variables as
in part B). Report your results in a new table. Interpret your results.

tes t score= + 0 small+ 1 aide+ 2 ( regular ) + 3 ( small )+ 4 ( aide ) + 5 ( otherregular ) + 6 ( othersmall )+

white in regular class =

white in small class =

white in class with aide =

Black in regular class =

Black in small class =

Black in class with aide =

Small Class
Regular with Aide
Black * Regular
Black * Small
Black * Aide
Other * Regular

(1)
Kindergarten
16.47***
(3.696)
1.403
(3.635)
-15.67**
(6.235)
-20.77***
(6.263)
-20.77***
(6.095)
-3.885

+ 0
+ 1
+ 2
+ 0 + 3
+ 1 + 4

(2)
Grade1
21.35***
(4.245)
10.02**
(4.257)
-20.75***
(6.816)
-15.74**
(6.733)
-35.97***
(7.577)
23.40

(3)
Grade2
12.25***
(4.262)
4.761
(4.279)
-33.50***
(7.102)
-29.67***
(6.423)
-40.18***
(6.969)
88.36**

(4)
Grade3
8.096**
(3.942)
0.479
(4.005)
-26.83***
(6.763)
-23.85***
(5.759)
-29.51***
(6.319)
40.76

Other * Small
Other * Aide
female
1st Quarter of Birth
2nd Quarter of Birth
3rd Quarter of Birth
Inner City School
Suburban School
Urban School
Free Lunch
N
R2

(33.06)
-3.660
(33.10)
-12.45
(38.13)
9.033***
(2.582)
-8.809**
(3.729)
-20.13***
(3.748)
-29.49***
(3.762)
-2.611
(7.513)
-8.409
(6.347)
-13.75**
(5.822)
-25.12***
(3.110)
2617
0.094

(34.16)
43.21
(38.18)
91.76*
(53.90)
5.934**
(2.939)
-9.246**
(4.264)
-17.38***
(4.291)
-26.30***
(4.278)
-39.50***
(8.551)
-17.23**
(7.267)
-7.061
(6.691)
-41.99***
(3.528)
2697
0.204

(43.10)
35.40
(33.42)
45.74
(43.09)
6.829**
(2.872)
-5.420
(4.161)
-10.33**
(4.184)
-18.45***
(4.172)
-18.17**
(8.361)
-10.40
(7.108)
1.630
(6.565)
-34.28***
(3.438)
2696
0.171

(38.64)
26.96
(29.93)
18.97
(38.60)
6.423**
(2.608)
-7.032*
(3.780)
-7.651**
(3.795)
-13.21***
(3.806)
-9.534
(7.583)
-5.296
(6.421)
-4.361
(5.911)
-33.16***
(3.160)
2629
0.136

(D) Write down the regression equation you would run to estimate whether
the effect of the two treatments is different for male versus female students.
For each grade level, run this regression (use the same control variables as
in part B). Report your results in a new table. Interpret your results.
tes t score= + 0 small+ 1 aide+ 2 ( femaleregular ) + 3 ( femalesmall ) + 4 ( femaleaide ) + X +

Male in regular class =

Male in small class =

Male in class with aide =

+ 1

Female in regular class =

+ 2

Female in small class =

Female in class with aide =

Small Class

(1)
Kindergarten
22.36***
(4.580)

+ 0

+ 0 + 3
+ 1 + 4

(2)
Grade1
26.44***
(5.175)

(3)
Grade2
14.24***
(5.128)

(4)
Grade3
11.29**
(4.854)

Regular + Aide
Female * Regular
Female * Small
Female * Aide
White
Black
Asian
Hispanic
1st Quarter of Birth
2nd Quarter of Birth
3rd Quarter of Birth
Inner City School
Suburban School
Urban School
Free Lunch
N
R2

2.610
(4.469)
15.25***
(4.397)
0.932
(4.595)
10.21**
(4.415)
-81.54*
(46.58)
-100.3**
(46.81)
-117.6**
(52.85)
-76.77
(65.76)
-8.360**
(3.724)
-19.80***
(3.742)
-29.20***
(3.759)
-3.640
(7.502)
-8.854
(6.332)
-14.32**
(5.811)
-24.86***
(3.107)
2617
0.097

1.717
(5.265)
5.657
(5.130)
-1.634
(4.916)
15.62***
(5.234)
-62.30
(53.83)
-85.00
(54.10)
-39.12
(61.09)
24.45
(76.02)
-8.745**
(4.262)
-17.12***
(4.284)
-25.91***
(4.276)
-39.96***
(8.540)
-17.04**
(7.262)
-7.077
(6.692)
-41.98***
(3.526)
2697
0.204

-1.029
(5.230)
5.047
(5.337)
2.750
(4.681)
12.80**
(4.967)
-121.4**
(52.68)
-155.5***
(52.93)
-79.24
(59.77)
-104.7
(74.40)
-5.064
(4.160)
-9.979**
(4.182)
-18.29***
(4.168)
-18.24**
(8.356)
-10.10
(7.103)
1.718
(6.557)
-33.92***
(3.437)
2696
0.172

-4.443
(4.953)
5.627
(5.201)
0.677
(4.108)
13.77***
(4.421)
-59.89
(47.17)
-86.27*
(47.40)
-46.69
(53.51)
-12.98
(66.61)
-6.605*
(3.778)
-7.328*
(3.790)
-12.99***
(3.802)
-9.893
(7.573)
-5.321
(6.413)
-4.564
(5.903)
-33.03***
(3.157)
2629
0.137

(E) Now suppose you wish to examine the effect of teacher experience on
student achievement, using the same outcome measure as in previous parts.
Teachers were randomly assigned to the three alternative class size
arrangements within schools, but not across schools.
Add teacher
experience as an additional control variable to the regressions estimated in
part B, and report the estimates in a table (for each grade level). Will the
estimated coefficients associated with the teacher experience variable
provide unbiased estimates of the impact of teacher experience on student
achievement? Explain.

(1)

(2)

(3)

(4)

Small Class
Regular + Aide
Teacher Experience
female
White
Black
Asian
Hispanic
1st Quarter of Birth
2nd Quarter of Birth
3rd Quarter of Birth
Inner City School
Suburban School
Urban School
Free Lunch
N
R2

Kindergarten
15.41***
(3.171)
-0.838
(3.109)
1.051***
(0.231)
8.680***
(2.569)
-80.56*
(46.41)
-99.27**
(46.64)
-116.1**
(52.64)
-70.00
(65.56)
-8.526**
(3.709)
-19.73***
(3.728)
-29.40***
(3.743)
-1.758
(7.475)
-9.901
(6.316)
-14.96***
(5.794)
-25.06***
(3.093)
2617
0.102

Grade1
22.26***
(3.568)
5.871
(3.684)
0.391**
(0.170)
6.236**
(2.936)
-65.60
(53.80)
-87.55
(54.07)
-44.02
(61.04)
26.08
(76.01)
-9.106**
(4.258)
-17.14***
(4.283)
-26.10***
(4.274)
-40.91***
(8.549)
-17.00**
(7.260)
-6.786
(6.694)
-42.19***
(3.525)
2697
0.204

Grade2
13.19***
(3.594)
3.716
(3.680)
-0.0396
(0.168)
6.864**
(2.886)
-123.7**
(52.71)
-158.1***
(52.96)
-81.85
(59.80)
-104.7
(74.48)
-5.536
(4.176)
-9.935**
(4.205)
-18.54***
(4.186)
-17.27**
(8.413)
-10.37
(7.123)
1.652
(6.567)
-33.77***
(3.449)
2674
0.167

Grade3
8.990***
(3.340)
-0.723
(3.433)
0.384**
(0.154)
6.722**
(2.617)
-59.49
(47.24)
-86.18*
(47.48)
-45.81
(53.60)
-5.422
(66.79)
-6.309*
(3.794)
-6.939*
(3.808)
-13.12***
(3.816)
-9.575
(7.623)
-5.143
(6.432)
-3.326
(5.930)
-33.18***
(3.178)
2613
0.137

(F) If the teacher experience estimates in part G are not necessarily


unbiased, are there covariates that could be included as additional controls
that would yield unbiased estimates of the effect of teacher experience on
student achievement? Explain your reasoning.
Yes, we can include teachers highest degree, teachers career ladder level,
and teachers race as other control variables.
(G) Add the additional covariates recommended in part H, and interpret the
results.
See the dofile and the document partG-Kindergarden Only.

(H) Finally, for each grade level, run a regression to examine whether having
a teacher that has the same race as the student has any impact on the
student achievement (include any control variables you deem appropriate for
this regression). Report your results in a table. Interpret your results.
See the dofile and the document partH-Kindergarden Only.
(I) [bonus question: adds 5% to your midterm]
It occurs to you that similarity of race between students and teachers might
have different impacts for different races. For example, while black students
might benefit from having a black teacher, white students might not benefit
to the same extent (or at all) from having a white teacher. Modify your
regression in part (J) to examine this new hypothesis. For each grade level,
run this new regression and report your results in a table. Interpret your
results.
See the dofile and the document partI-Kindergarden Only.

You might also like