Professional Documents
Culture Documents
Joshua Angrist
MIT and NBER
Outcome data
Collected mostly from university systems (GPA, credits, academic
probation, retention)
Outcomes collected at the end of rst semester, rst year, second year
We surveyed a subset later, treatment and control (low response)
Focus groups with a few treated students (they loved the program!)
Take-up (sign-up, participation)
Many students who were oered services were uninterested; SSP
sign-up and service use rates were low
Fellowship treatments boosted sign-up
Women signed up more than men
STAR Results , , , (rst year)
Services alone have no eect
Incentives wear o quickly
Top marks for both (at least for women)
These results are mixed; Our follow-up is downright disappointing!
Do You Believe in Parallel Worlds?
Meat and Potatoes
On the menu of identication strategies, Ill usually take IV, RD, and
DD ... in that order
Dierences-in-dierences is our staple, meat and potatoes while were
waiting for something spicier
Lotteries that creates instruments, perhaps
An interesting discontinuity
Yet, dis-in-dis can be compelling ... sometimes it makes a nice
shepherds pie
At the LSE, I had to learn to eat such things (and much worse!)
Alas, you say, only the Labor tribe can partake
Not so: heres a DD banking story for our times
The story may be familiar to you: I recently found this artifact in the
British Museum
Otherwise, mostly boring stu there
Banking on the South
In the late 1920s, Caldwell and Company ran the American Souths
largest bank chain and owned many other businesses
In November 1930, corruption, mismanagement, and drought
collapsed the Caldwell empire
Within 6 months, 29 Mississippi banks folded as depositors panicked
Does this sort of nancial collapse necessarily precipitate a decline in
economic activity?
Policy makers facing a bank run can let credit ow easy or turn o
the tap:
Easy money allows troubled banks to meet withdrawal demands,
staving o depositor panic, and obviating the need for support in the
rst place
Are all liquidity crises merely crises of condence? Bagehot (1873): If
the banks are bad, they will certainly continue bad and will probably
become worse if the Government sustains and encourages them ... aid
to a present bad bank is the surest mode of preventing the
establishment of a future good bank.
One Mississippi, Two Mississippi (based on a story by
Richardson and Troost)
log yd ,t = d + t
Yd ,t = + TREATd + POSTt
+rDD (TREATd POSTt ) + ed ,t
50
Vertical lines mark two
standard errors
40
30
Log Points
20
10
-10
-20
2 Years 1 Year Year of 1 Year 2 Years 3 Years 4 or More
Prior Prior Adoption After After After Years After
Time relative to adoption of implied contract exception
Table 5.2.3
Estimated effects of labor regulation on the performance of firms
in Indian states
(1) (2) (3) (4)
Labor regulation (lagged) .186 .185 .104 .0002
(.064) (.051) (.039) (.020)
Log development .240 .184 .241
expenditure per capita (.128) (.119) (.106)
Log installed electricity .089 .082 .023
capacity per capita (.061) (.054) (.033)
Log state population .720 0.310 1.419
(.96) (1.192) (2.326)
Congress majority .0009 .020
(.01) (.010)
Hard left majority .050 .007
(.017) (.009)
Janata majority .008 .020
(.026) (.033)
Regional majority .006 .026
(.009) (.023)
State-specific trends No No No Yes
Adjusted R2 .93 .93 .94 .95
Notes: Adapted from Besley and Burgess (2004), table IV. The table reports
regression DD estimates of the effects of labor regulation on productivity. The
dependent variable is log manufacturing output per capita. All models include
state and year effects. Robust standard errors clustered at the state level are
Getting a Little Jumpy -
The Regression Discontinuity Idea
1 if xi x0
di = . (3)
0 if xi < x0
AL-99 exploit the fact that Israeli class size is capped at 40; if
enrollment exceeds this threshold, classes are split
This suggests we compare, say, kids in schools with 38-40 fth
graders to kids in schools with 41-43 fth graders, where classes are
much smaller
The idea is that this small dierence in enrollment, which generates a
big change in class size, is a "good experiment"
Let msc denote the predicted class size (in a given grade) assigned to
class c in school s, where enrollment in the grade is denoted es .
Maimonides Rule is
es
msc = (e s 1 )
int [ 40 ] + 1
TABLE II
OLS ESTIMATES FOR 1991
5th Grade
The unit of observation is the average score in the class. Standard errors are reported in parentheses. Standard errors were cor
class eect of about 5 to 6 percentile points. The eect size is about :2 ; where is the standard deviation
of the
- percentile score in kindergarten. The small-class eect is signicantly dierent from zero, while the
Table 2.2.2: Experimental estimates of the eect of class-size assignment on test scores
Table 4.4.1
Results from the JTPA experiment: OLS and IV estimates of training impacts
Comparisons by Comparisons by Instrumental Variable
Training Status (OLS) Assignment Status (ITT) Estimates (IV)
Without With Without With Without With
Covariates Covariates Covariates Covariates Covariates Covariates
(1) (2) (3) (4) (5) (6)
A. Men 3,970 3,754 1,117 970 1,825 1,593
(555) (536) (569) (546) (928) (895)
B. Women 2,133 2,215 1,243 1,139 1,942 1,780
(345) (334) (359) (341) (560) (532)
Notes: Authors tabulation of JTPA study data. The table reports OLS, ITT, and IV
estimates of the effect of subsidized training on earnings in the JTPA experiment. Columns 1
and 2 show differences in earnings by training status; columns 3 and 4 show differences by
random-assignment status. Columns 5 and 6 report the result of using random-assignment
status as an instrument for training. The covariates used for columns 2, 4, and 6 are high
school or GED, black, Hispanic, married, worked less than 13 weeks in past year, AFDC
(for women), plus indicators for the JTPA service strategy recommended, age group, and
second follow-up survey. Robust standard errors are shown in parentheses. There are 5,102
men and 6,102 women in the sample.
-
Signed up for STAR Received SSP services Met with/emailed an advisor Attended FSGs
Basic All Basic All Basic All Basic All
controls controls controls controls controls controls controls controls
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. All
Offered SSP 0.519 0.549 0.238 0.255 0.204 0.217 0.106 0.118
[0.032]*** [0.034]*** [0.028]*** [0.029]*** [0.026]*** [0.028]*** [0.020]*** [0.021]***
Offered SFP 0.863 0.867
[0.022]*** [0.022]***
Offered SSP 0.762 0.792 0.412 0.431 0.383 0.397 0.131 0.139
and SFP [0.036]*** [0.036]*** [0.041]*** [0.044]*** [0.041]*** [0.043]*** [0.029]*** [0.031]***
Observations 1,571 1,431 1,571 1,431 1,571 1,431 1,571 1,431
Panel B. Men
Offered SSP 0.447 0.464 0.194 0.206 0.145 0.149 0.096 0.107
[0.049]*** [0.052]*** [0.039]*** [0.042]*** [0.035]*** [0.038]*** [0.029]*** [0.032]***
Offered SFP 0.792 0.806
[0.040]*** [0.040]***
Offered SSP 0.705 0.708 0.298 0.291 0.282 0.270 0.115 0.112
and SFP [0.058]*** [0.065]*** [0.058]*** [0.063]*** [0.057]*** [0.061]*** [0.042]*** [0.046]**
Observations 665 594 665 594 665 594 665 594
Panel C. Women
Offered SSP 0.571 0.605 0.273 0.287 0.251 0.264 0.113 0.124
[0.043]*** [0.044]*** [0.038]*** [0.040]*** [0.037]*** [0.040]*** [0.027]*** [0.029]***
Offered SFP 0.912 0.908
[0.024]*** [0.026]***
Offered SSP 0.800 0.835 0.506 0.532 0.466 0.489 0.146 0.155
and SFP [0.046]*** [0.043]*** [0.056]*** [0.058]*** [0.056]*** [0.058]*** [0.040]*** [0.042]***
Observations 906 837 906 837 906 837 906 837
-
Table 5Treatment Effects on First Year Outcomes in the Sample with Fall Grades
Notes: The table reports regression estimates of treatment effects on full grades and first-year GPA compu
-
Our basic econometric model is FIGURE 1.STATE LOG EMPLOYMENT-TO-POPULATION RATIOS BEFORE AND
AFTER ADOPTION OF IMPLIED-CONTRACT EXCEPTION: MONTHLY LEADS AND
LAGS FROM 4 YEARS BEFORE TO 8 YEARS AFTER ADOPTION
Y st 1 Treat st 2 Post st
(1)
3TreatstPostst st,
40
30
Class size
20
10
A. Fifth Grade
0
40
30
Class size
20
10
B. Fourth Grade
0
makes the RD design fuzzy. Still, there are clear drops in class
-
Table 6.2.1
OLS and fuzzy RD estimates of the effect of class size on
fth-grade math scores
OLS 2SLS
Full Sample Discontinuity Samples
5 3
(1) (2) (3) (4) (5) (6) (7) (8)
Mean score 67.3 67.3 67.0 67.0
(SD) (9.6) (9.6) (10.2) (10.6)
Regressors
Class size .322 .076 .019 .230 .261 .185 .443 .270
(.039) (.036) (.044) (.092) (.113) (.151) (.236) (.281)
Percent .340 .332 .350 .350 .459 .435
disadvantaged (.018) (.018) (.019) (.019) (.049) (.049)
Enrollment .017 .041 .062 .079
(.009) (.012) (.037) (.036)
Enrollment .010
squared/100 (.016)
Segment 1 12.6
(enrollment 3843) (3.80)
Segment 2 2.89
(enrollment 7883) (2.41)
R2 .048 .249 .252
Number of classes 2,018 2,018 471 302
Notes: Adapted from Angrist and Lavy (1999). The table reports estimates of equation
(6.2.6) in the text using class averages. Standard errors, reported in parentheses, are corrected
for within-school correlation.
- 254 Chapter 6
A. LINEAR E[Y0i|Xi]
1.5
1.0
Outcome
0.5
0
0 0.2 0.4 0.6 0.8 1.0
X
B. NONLINEAR E[Y0i|Xi]
1.5
1.0
Outcome
0.5
1.0
Outcome
0.5
-0.5
0 0.2 0.4 0.6 0.8 1.0
X
8th District
160
Number of Banks in Business
140
6th District
120
Treatment effect
180 160
Number of Banks in Business
8th District
140
6th District
120 100
8th District
Number of Banks in Business
140
6th District
100 120
-
1066 journal of political econom
TABLE 8
Decline in Wholesale Trade