Raising the Bar: Impacts and Implementation of the New Heights Program for Expectant and Parenting Teens in Washington, dc

Table C.2. New Heights impacts under alternative approaches to calculating 
standard errors 
. 
Unexcused 
absences 
per 
semester
Excused 
absences 
per 
semester 
Days 
attended 
per 
semester
Credits 
earned 
per year 
Semester 
graduation 
rate
a 
(percentage) 
Sample 
size 
Main analysis results School 
fixed effects, standard errors 
clustered at student level 
-4.54
***
1.35
**
3.43
***
1.06
***
0.03
*
10,760 
Standard errors clustered at 
school level
b
-4.54
1.35
**
3.43
1.06
**
0.03
10,760 
Standard errors clustered at the 
semester level
b
-4.54
***
1.35
***
3.43
***
1.06
***
0.03
***
10,760 
Standard errors clustered at the 
school-by-semester level
b
-4.54
***
1.35
3.43
**
1.06
***
0.03
*
10,760 
Mixed effects models 
. 
. 
. 
. 
. 
. 
School fixed effects, student 
random effects 
-6.61
***
0.79
7.29
***
1.30
***
0.03
*
10,760 
School random effects
c
-4.54
1.35
**
3.43
1.06
**
0.03
10,760 
School fixed effects, semester 
random effects 
-4.52
***
1.35
3.41
***
1.06
***
0.03
10,760 
School fixed effects, semester-
by-school random effects 
-3.42
**
1.54
2.39
0.65
***
0.01
10,760 
Source: DCPS administrative date; DC DOH administrative data. 
Note: 
All regressions, unless otherwise noted, include time period fixed effects (year for credits earned per year, and 
semester for all other outcomes); age indicators; race and ethnicity indicators; and an indicator for being over-age 
when entering 9th grade. 
P
-values are based on standard errors made robust to serial correlation within students, 
unless otherwise noted. Statistical significance is based on a two-tailed t-test. Statistical significance within the 
three outcomes in the attendance domain, unexcused absences, excused absences, and days attended are 
based on 
p
-values that are adjusted for multiple comparisons using a Bonferroni adjustment. 
a
The semester graduation rate is the proportion of students who graduate each semester. The analysis is restricted to 
students who are at least 17 years old each semester. The semester graduation rate is a marginal measure of graduation, 
whereas the more commonly used cohort graduation rate is a cumulative measure. The semester graduation rate is lower 
than the cohort graduation rate. 
b 
Changing the level of clustering affects only how the standard errors are calculated, whereas the point estimate is 
algebraically unchanged.
c
Modeling school effects as fixed or random has no effect on the point estimate (to two decimal places) due to the large size 
of schools (Wooldridge 2010
25
). 
*Significantly different from zero at the .10 level. 
**Significantly different from zero at the .05 level. 
25
Wooldridge, Jeffrey M. 
Econometric Analysis of Cross Section and Panel Data
. Cambridge, MA: MIT Press, 
2010. 
C.7 

---

APPENDIX C: ANALYTIC METHODS OF IMPACT ANALYSIS AND RESULTS FROM ALTERNATIVE MODELS
 
***Significantly different from zero at the .01 level.