Joshua Angrist Nobel Prize Lecture 2021
-
0:01 - 0:04♪ [music] ♪
-
0:11 - 0:13- [Joshua] As I stilled
my trembling iPhone -
0:13 - 0:14early on October 11th,
-
0:14 - 0:18my thoughts went to the question
of whether Nobel-level recognition -
0:18 - 0:21might change life
for the Angrist family. -
0:21 - 0:23Ours is a close-knit family.
-
0:23 - 0:24We lack for nothing.
-
0:24 - 0:27So I worried
that stressful Nobel celebrity -
0:27 - 0:28might not be a plus.
-
0:28 - 0:30But with the first cup of coffee,
-
0:30 - 0:32I began to relax.
-
0:32 - 0:33It occurred to me
-
0:33 - 0:35that the matter
of how public recognition -
0:35 - 0:37affects a scholar's life
-
0:37 - 0:40is, after all,
a simple causal question. -
0:40 - 0:42The Nobel intervention
-
0:42 - 0:45is substantial, sudden,
and well-measured. -
0:45 - 0:48Outcomes like health and wealth
are easy to record. -
0:49 - 0:52Having just been recognized
with my co-laureates, -
0:52 - 0:54Guido Imbens and David Card,
-
0:54 - 0:57for answering causal questions
using observational data, -
0:57 - 1:00my thoughts moved
from personal upheaval -
1:00 - 1:02to the more familiar demands
-
1:02 - 1:06of identification and estimation
of causal effects. -
1:06 - 1:08I was able to soothe
my worried mind -
1:08 - 1:13by imagining a study
of the Nobel Prize treatment effect. -
1:13 - 1:15How would such a study
be organized? -
1:16 - 1:20In a 1999 essay published in
the "Handbook of Labor Economics," -
1:20 - 1:24Alan Krueger and I embraced
the phrase "empirical strategy." -
1:24 - 1:26The handbook volume
in question was edited -
1:26 - 1:29by two of my Princeton
Ph.D. thesis advisors, -
1:29 - 1:32Orley Ashenfelter and David Card --
-
1:32 - 1:35among the most successful
and prolific graduate advisors -
1:35 - 1:37economics has known.
-
1:37 - 1:39An empirical strategy
is a research plan -
1:39 - 1:41that encompasses data collection,
-
1:41 - 1:44identification,
and econometric estimation. -
1:45 - 1:48Identification is the applied
econometricians term -
1:48 - 1:50for research design --
-
1:50 - 1:52a randomized clinical trial,
-
1:52 - 1:56an RCT, is the simplest
and most powerful research design. -
1:57 - 1:59In RCTs, causal effects
are identified -
1:59 - 2:02by the random assignment
of treatment. -
2:02 - 2:03Random assignment ensures
-
2:03 - 2:05that treatment and control groups
-
2:05 - 2:07are comparable
in the absence of treatment. -
2:07 - 2:09So differences
between them afterwards -
2:09 - 2:12reflect only the treatment effect.
-
2:12 - 2:15Nobel Prizes are probably
not randomly assigned. -
2:15 - 2:17This challenge notwithstanding,
-
2:17 - 2:20a compelling empirical strategy
for the Nobel treatment effect -
2:20 - 2:21comes to mind,
-
2:21 - 2:24at least as a flight
of empirical fancy. -
2:24 - 2:28Imagine a pool of prize-eligible
Nobel applicants, -
2:28 - 2:31the group under consideration
for the prize. -
2:31 - 2:33Applicants need not
apply themselves. -
2:33 - 2:36They are, I presume, nominated
by their peer scholars. -
2:37 - 2:41My fanciful Nobel impact study
looks only at Nobel applicants, -
2:41 - 2:43since these are all elite scholars.
-
2:43 - 2:45But that is only the first step.
-
2:45 - 2:48Credible applicants, I imagine,
are evaluated by judges -
2:48 - 2:54using criteria like publications,
citations, nominating statements. -
2:54 - 2:58I imagine this material is reviewed
and assigned a numerical score, -
2:58 - 3:00using some kind of scoring rubric.
-
3:01 - 3:04Top scorers up to three per field,
in any single year -
3:04 - 3:06win a prize.
-
3:06 - 3:08Having identified applicants
-
3:08 - 3:10and collected data
on their scores, -
3:10 - 3:12the next step
in my Nobel impact study -
3:12 - 3:15is to record the relevant cutoffs.
-
3:15 - 3:17The Nobel cutoff
is the lowest score -
3:17 - 3:19among those awarded a prize.
-
3:20 - 3:23Many Nobel hopefuls
just missed the cutoff. -
3:23 - 3:27Looking only at near misses,
along with the winners, -
3:27 - 3:28differences in scores
-
3:28 - 3:30between those above
and below the cutoff -
3:30 - 3:35begin to look serendipitous,
almost randomly assigned. -
3:35 - 3:36After all,
-
3:36 - 3:39near-Nobels are among
the most eminent of scholars, too. -
3:40 - 3:43With one more
high-impact publication, -
3:43 - 3:45a little more support
from nominators, -
3:45 - 3:46they would have been awarded
-
3:46 - 3:47Nobel gold.
-
3:47 - 3:50Some of them, someday,
surely will be. -
3:50 - 3:52The empirical strategy
sketched here is called -
3:52 - 3:54a Regression Discontinuity Design,
-
3:54 - 3:56RD.
-
3:56 - 3:58RD exploits the jumps
in human affairs, -
3:58 - 4:01induced by rules, regulations,
-
4:01 - 4:05and the need to classify people
for various assignment purposes. -
4:05 - 4:07When treatment
or intervention is determined -
4:07 - 4:10by whether a tiebreaking variable
crosses a threshold, -
4:10 - 4:13those just below the threshold
become a natural control group -
4:13 - 4:15for those who clear it.
-
4:15 - 4:17RD does not require
that the variable -
4:17 - 4:19whose causes we seek,
-
4:19 - 4:22switch fully on or fully off
at the cutoff. -
4:22 - 4:23We require only
-
4:23 - 4:26that the average value
of this variable -
4:26 - 4:27jump at the cutoff.
-
4:27 - 4:30RD can allow,
for example, for the fact -
4:30 - 4:33that this year's near-Nobel
might be next year's winner. -
4:33 - 4:36Allowing for this
leads to the use of jumps -
4:36 - 4:38in the rate at which
treatment is assigned -
4:38 - 4:41to construct
instrumental variables, IV, -
4:41 - 4:44estimates of the effect
of treatment received. -
4:44 - 4:47This sort of RD
is said to be "fuzzy," -
4:47 - 4:50But as Steve Pischke and I
wrote in our first book: -
4:50 - 4:52"Fuzzy RD is IV."
-
4:52 - 4:53[kids cheering]
-
4:53 - 4:55The first RD study I contributed to
-
4:55 - 4:58was written with my frequent
collaborator, Victor Lavy. -
4:58 - 4:59This study is motivated
-
4:59 - 5:02by the high costs
and uncertain returns -
5:02 - 5:04to smaller elementary school classes.
-
5:04 - 5:07We exploited a rule
used by Israeli elementary schools -
5:07 - 5:09to determine class size.
-
5:09 - 5:12This rule is used to estimate
class-size effects, -
5:12 - 5:15as if in a class-size RCT.
-
5:16 - 5:19In the 1990s,
Israeli classes were large. -
5:19 - 5:22Students enrolled
in a grade cohort of 40 -
5:22 - 5:25were likely to be seated
in a class of 40 -- -
5:25 - 5:27that's the relevant cutoff.
-
5:27 - 5:30Add another child
to the cohort, making 41, -
5:30 - 5:32and the cohort
was likely to be split -
5:32 - 5:35into two much smaller classes.
-
5:35 - 5:39This leads to the Maimonides'
rule research design, -
5:39 - 5:41so named because
the 12th-century Rambam -
5:41 - 5:44proposed a maximum
class size of 40. -
5:44 - 5:47This figure plots Israeli
4th grade class sizes -
5:47 - 5:50as a function
of 4th grade enrollment, -
5:50 - 5:52overlaid with the theoretical
class size rule, -
5:53 - 5:54Maimonides' rule.
-
5:54 - 5:56The fit isn't perfect --
-
5:56 - 5:59that's a feature that makes
this application of RD fuzzy. -
5:59 - 6:02But the gist of the thing
is a marked class size drop -
6:02 - 6:06at each integer multiple of 40,
the relevant cutoff, -
6:06 - 6:08just as predicted by the rule.
-
6:08 - 6:11As it turns out,
these drops in class size -
6:11 - 6:12are reflected in jumps
-
6:12 - 6:15in 4th and 5th grade math scores.
-
6:20 - 6:23Would a comparison
of Nobel laureates to near-laureates -
6:23 - 6:25really be a good
natural experiment? -
6:26 - 6:28The logic behind this sort of claim
-
6:28 - 6:30seems more compelling
for comparisons of schools -
6:30 - 6:32with 40 and 41 fourth graders
-
6:33 - 6:36than for comparisons
of laureates and near-laureates. -
6:36 - 6:40Yet, both scenarios exploit
a feature of the physical world. -
6:40 - 6:42Provided the tiebreaking variable,
-
6:42 - 6:45known to RD mavens
as the "running variable," -
6:45 - 6:47has a continuous distribution,
-
6:47 - 6:51the probability of crossing
the cutoff approaches one half -
6:51 - 6:52when examined in a narrow window
-
6:52 - 6:54around the cutoff.
-
6:54 - 6:56In RD empirical work,
-
6:56 - 6:59the window around such cutoffs
is known as a bandwidth. -
7:00 - 7:04Importantly, this limiting
probability is 0.5 for everybody -
7:04 - 7:06regardless of
how qualified they look -
7:06 - 7:08going into the Nobel competition.
-
7:08 - 7:12This remarkable fact can be seen
in data on applicants -
7:12 - 7:15to one of New York's
highly coveted screen schools. -
7:15 - 7:16By way of background,
-
7:16 - 7:20roughly 40% of New York City's
middle and high schools -
7:20 - 7:24select their applicants
on the basis of test scores, grades, -
7:24 - 7:26and other exacting criteria.
-
7:26 - 7:27In other words,
-
7:27 - 7:29the admissions regime
for screen schools -
7:29 - 7:31is a lot like the scheme
I've imagined -
7:31 - 7:33for the Nobel Prize.
-
7:33 - 7:35Screen schools
are but one of a number -
7:35 - 7:37of highly selective systems
within a system -
7:37 - 7:40in large U.S. school districts.
-
7:40 - 7:43Boston, Chicago, San Francisco,
and Washington, D.C. -
7:43 - 7:46all feature highly selective
institutions, -
7:46 - 7:48often known as exam schools.
-
7:48 - 7:49Exam schools operate
-
7:49 - 7:51as part of larger
public school systems -
7:51 - 7:54that enroll students
without screening. -
7:54 - 7:56Motivated by
the enduring controversy -
7:56 - 7:58over the equity
of screened admissions, -
7:58 - 8:00my Blueprint
Labs collaborators and I -
8:00 - 8:04have examined the causal effects
of exam school attendance -
8:04 - 8:06in Boston, Chicago, and New York.
-
8:06 - 8:09This figure shows the probability
of being offered a seat -
8:09 - 8:12at New York's storied
Townsend Harris High School, -
8:12 - 8:15ranked 12th nationwide.
-
8:15 - 8:18Bar height in the figure
marks the qualification rate -- -
8:18 - 8:22that is, the likelihood of earning
a Townsend Harris admission score -
8:22 - 8:26above that of the lowest
scoring applicant offered a seat. -
8:26 - 8:30Importantly, the bars show
qualification rates conditional -
8:30 - 8:34on a measure of pre-application
baseline achievement. -
8:34 - 8:37In particular, the bars mark
qualification rates conditional -
8:37 - 8:38on whether an applicant
-
8:38 - 8:43has upper quartile or lower quartile
6th grade math scores. -
8:43 - 8:46Townsend Harris applicants
with high baseline scores -
8:46 - 8:48are much more likely to qualify
-
8:48 - 8:50than applicants
with low baseline scores. -
8:50 - 8:52This isn't surprising.
-
8:52 - 8:55But in a shrinking
symmetric bandwidth -
8:55 - 8:57around the school's cutoff,
-
8:57 - 9:00qualification rates
in the two groups converge. -
9:00 - 9:04Qualification rates in the last
and smallest groups -
9:04 - 9:06are both remarkably close to one half.
-
9:07 - 9:08This is what we'd expect to see
-
9:08 - 9:10were Townsend Harris
to admit students -
9:10 - 9:12by tossing a coin,
-
9:12 - 9:15rather than by selecting
only those who scored highly -
9:15 - 9:17on the school's entrance exam.
-
9:17 - 9:20Even when admissions
operates by screening, -
9:20 - 9:23the data can be arranged
so as to mimic an RCT. -
9:28 - 9:31Few of the questions I've studied
are more controversial -
9:31 - 9:34than the question of access
to public exam schools, -
9:34 - 9:36like the Boston Latin School,
-
9:36 - 9:39Chicago's Payton and Northside
selective enrollment high schools, -
9:39 - 9:42and New York's legendary
Brooklyn Tech, -
9:42 - 9:43Bronx Science,
-
9:43 - 9:46and Stuyvesant
specialized high schools, -
9:46 - 9:49which have graduated
14 Nobel laureates between them. -
9:49 - 9:52Townsend Harris, the school
we started with today, -
9:52 - 9:56graduated three Nobels,
including economist Ken Arrow. -
9:56 - 9:57Exam school proponents
-
9:57 - 9:59see the opportunities
these schools provide -
9:59 - 10:02as democratizing public education.
-
10:02 - 10:04Wealthy families, they argue,
-
10:04 - 10:07can access exam school curricula
in the private sector. -
10:08 - 10:10Shouldn't ambitious
low-income students -
10:10 - 10:12be afforded the same chance
at elite education? -
10:13 - 10:15Critics of selective
enrollment schools -
10:15 - 10:18argue that rather
than expanding equity, -
10:18 - 10:20exam schools are inherently biased
-
10:20 - 10:22against the Black
and Hispanic students -
10:22 - 10:25that make up the bulk
of America's urban districts. -
10:25 - 10:28New York's super selective
Stuyvesant, for example, -
10:28 - 10:31enrolled only
seven Black students in 2019, -
10:31 - 10:34out of an incoming class of 895.
-
10:34 - 10:38But are exam school seats
really worth fighting for? -
10:39 - 10:42My collaborators and I
have repeatedly used -
10:42 - 10:45RD empirical strategies
to study the causal effects -
10:45 - 10:47of attendance at exam schools
-
10:47 - 10:49like Townsend Harris
and Boston Latin. -
10:49 - 10:51Our first exam school study,
-
10:51 - 10:54which looks at schools
in Boston and New York, -
10:54 - 10:56encapsulates
these findings in its title: -
10:56 - 10:57"The Elite Illusion."
-
10:58 - 11:00The elite illusion
refers to the fact -
11:00 - 11:04that while exam school students
undoubtedly have high test scores -
11:04 - 11:05and other good outcomes,
-
11:05 - 11:08this is not a causal effect
of exam school attendance. -
11:09 - 11:11Our estimates consistently suggest
-
11:11 - 11:14that the causal effects
of exam school attendance -
11:14 - 11:17on their students learning
and college-going are zero -- -
11:17 - 11:20maybe even negative.
-
11:20 - 11:22The good performance
of exam school students -
11:22 - 11:24reflect selection bias --
-
11:24 - 11:27that is, the process by which
these students are chosen, -
11:27 - 11:29rather than causal effects.
-
11:30 - 11:32Data from Chicago's large exam
school sector -
11:32 - 11:34illustrate the elite illusion.
-
11:34 - 11:37This figure plots
peer mean achievement -- -
11:37 - 11:42that is, the 6th grade test scores
of my 9th grade classmates -
11:42 - 11:44against the admissions tiebreaker
-
11:44 - 11:46for a subset of applicants
-
11:46 - 11:49to any one of Chicago's
nine exam schools. -
11:49 - 11:52Applicants to these schools
rank up to 6, -
11:52 - 11:54while the exam schools
prioritize their applicants -
11:54 - 11:57using a common composite index,
-
11:57 - 11:59formed from an admissions test,
-
11:59 - 12:02GPAs, and grade 7
standardized scores. -
12:03 - 12:06This composite tiebreaker
is the running variable -
12:06 - 12:09for an RD design
that reveals what happens -
12:09 - 12:12when any applicant is offered
an exam school seat. -
12:12 - 12:14In Chicago's exam school match,
-
12:14 - 12:16which is actually an application
-
12:16 - 12:19of the celebrated Gale and Shapley
matching algorithm, -
12:19 - 12:22exam school applicants are sure
to be offered a seat somewhere -
12:23 - 12:26when they clear the lowest
in their set of cutoffs -
12:26 - 12:28among the schools they rank.
-
12:28 - 12:31We call this lowest cutoff
the "qualifying cutoff." -
12:32 - 12:35The figure shows a sharp jump
in peer mean achievement -
12:35 - 12:37for Chicago exam school applicants
-
12:37 - 12:39who clear their qualifying cutoff.
-
12:39 - 12:41This jump reflects the fact
-
12:41 - 12:44that most applicants
offered an exam school seat take it, -
12:44 - 12:46and applicants who enroll
-
12:46 - 12:48at one of Chicago's selective
enrollment high schools -
12:48 - 12:51are sure to be seated
in a 9th grade classroom -
12:51 - 12:53filled with academically
precocious peers, -
12:53 - 12:57because only the relatively
precocious make it in. -
12:57 - 13:00The increase in peer achievement
across the qualifying cutoff -
13:00 - 13:03amounts to almost
half a standard deviation -- -
13:03 - 13:04a very large effect.
-
13:04 - 13:07And yet, precocious peers
notwithstanding, -
13:07 - 13:09the offer of an exam school seat
-
13:09 - 13:11does not appear
to increase learning. -
13:12 - 13:16Let's plot applicants ACT scores
against their tiebreaker values. -
13:16 - 13:19This plot shows that exam
school applicants -
13:19 - 13:21who clear their qualifying cutoff
-
13:21 - 13:24perform sharply worse on the ACT.
-
13:24 - 13:25What explains this?
-
13:26 - 13:29It takes a tale of IV and RD
to untangle the forces -
13:29 - 13:33behind this intriguing
and unexpected negative effect. -
13:33 - 13:35But first, some IV theory.
-
13:40 - 13:43Guido Imbens and I
developed theoretical tools -
13:43 - 13:45that enhance
economists' understanding -
13:45 - 13:46of empirical strategies
-
13:46 - 13:48involving IV and RD.
-
13:49 - 13:52The prize we share
is in recognition of this work. -
13:52 - 13:55Guido and I overlapped
for only one year at Harvard, -
13:55 - 13:58where we had both taken
our first jobs post Ph.D. -
13:58 - 14:01I welcomed Guido
to Cambridge, Massachusetts -
14:01 - 14:04with a pair of interesting
instrumental variables. -
14:04 - 14:06I had used
the draft lottery instrument -
14:06 - 14:08in my Ph.D. thesis
-
14:08 - 14:11to estimate the long-run
economic consequences -
14:11 - 14:13of serving in the Armed Forces
-
14:13 - 14:15for soldiers who were drafted.
-
14:15 - 14:17The draft lottery instrument
relies on the fact -
14:17 - 14:20that lottery numbers
randomly assigned to birthdays -
14:20 - 14:23determined Vietnam-era
conscription risk. -
14:24 - 14:26Yet, even then, most soldiers
were volunteers, -
14:26 - 14:28as they are today.
-
14:28 - 14:29The quarter birth instrument
-
14:29 - 14:32is used in my 1991 paper
with Alan Krueger -
14:32 - 14:35to estimate the economic
returns to schooling. -
14:35 - 14:36This instrument uses the fact
-
14:36 - 14:39that men who were born
earlier in the year -
14:39 - 14:40were allowed to drop out
of high school -
14:40 - 14:42on their 16th birthday
-
14:42 - 14:45with less schooling completed
than those born later. -
14:45 - 14:48Guido and I soon began
asking each other, -
14:48 - 14:51"What really do we learn
from the draft eligibility -
14:51 - 14:53and quarter of birth
natural experiments?" -
14:54 - 14:57An early result in our quest
for a new understanding of IV -
14:57 - 15:00was a solution to the problem
of selection bias -
15:00 - 15:02in an RCT with partial compliance.
-
15:03 - 15:05Even in a randomized clinical trial,
-
15:05 - 15:08some of the people assigned
to treatment may opt out. -
15:08 - 15:11This fact has long vexed trialists
-
15:11 - 15:15because decisions to opt out
are not made by random assignment. -
15:15 - 15:17Our first manuscript together
-
15:17 - 15:21shows that in a randomized trial
with partial compliance, -
15:21 - 15:22you can use IV
-
15:22 - 15:24to estimate the effect
of treatment on the treated, -
15:24 - 15:26even when some offered treatment
-
15:26 - 15:27decline it.
-
15:27 - 15:29This works in spite of the fact
-
15:29 - 15:30that those who comply
with treatment -
15:30 - 15:33may be a very select group.
-
15:33 - 15:36Unfortunately, for us,
we were late to the party. -
15:36 - 15:39Not long after releasing
our first working paper, -
15:39 - 15:42we learned of a concise contribution
from Howard Bloom -
15:42 - 15:44that includes this theoretical result.
-
15:44 - 15:48Remarkably, Bloom had derived
this from first principles -
15:48 - 15:50without making a connection to IV.
-
15:50 - 15:52So Guido and I went back
to the drawing board. -
15:52 - 15:55And a few months later,
we had LATE -- -
15:55 - 15:56a theorem showing how to estimate
-
15:56 - 15:59the Local Average Treatment Effect.
-
15:59 - 16:02The LATE theorem
generalizes the Bloom theorem -
16:02 - 16:06and establishes the connection
between compliance and IV. -
16:06 - 16:08Maintaining the clinical
trials analogy, -
16:08 - 16:12let "Zi" indicate whether subject "i"
is offered treatment. -
16:12 - 16:13This is randomly assigned.
-
16:13 - 16:17Also, let "D1i" indicate
subject i's treatment status -
16:17 - 16:18when assigned to treatment,
-
16:18 - 16:21and let "D0i" indicate
subject i's treatment status -
16:21 - 16:23when assigned to control.
-
16:23 - 16:25I'll use this formal notation
-
16:25 - 16:27to give a clear statement
of the LATE result, -
16:27 - 16:29and then follow up with examples.
-
16:30 - 16:31A key piece of the LATE framework,
-
16:31 - 16:34pioneered by statistician Don Rubin,
-
16:34 - 16:36is the pair of potential outcomes.
-
16:36 - 16:38As is customary,
-
16:38 - 16:40I denote potential outcomes
for subject i -
16:40 - 16:42in the treated and untreated states
-
16:42 - 16:45by "Y1i" and "Y0i", respectively.
-
16:46 - 16:49The observed outcome
is Y1i for the treated -
16:49 - 16:51and Y0i for those not treated.
-
16:52 - 16:54Y1i minus Y0i
-
16:54 - 16:57is the causal effect
of treatment on individual i, -
16:57 - 16:59but this we can never see.
-
16:59 - 17:03We try, therefore, to estimate
some kind of average causal effect. -
17:03 - 17:06The LATE framework
allows us to do that in an RCT -
17:06 - 17:08where some controls are treated.
-
17:08 - 17:09The theorem says
-
17:09 - 17:10that the average causal
effect on people -
17:10 - 17:12whose treatment status
can be changed -
17:12 - 17:14by the offer of treatment
-
17:14 - 17:17is the ratio of ITT
to the treatment control difference -
17:17 - 17:18in compliance rates.
-
17:19 - 17:21A mathematical statement
of this result appears here, -
17:22 - 17:26where Greek letter Delta
symbolizes the ITT effect -
17:26 - 17:29and Greek symbols Pi1 and Pi0
-
17:29 - 17:31are compliance rates
in the group assigned to treatment -
17:31 - 17:34and the group assigned
to control, respectively. -
17:35 - 17:36The print version of this lecture
-
17:36 - 17:39delves deeper
into LATE intellectual history, -
17:39 - 17:41highlighting key contributions
made with Rubin. -
17:42 - 17:45For now, though, I'd like to make
the LATE theorem concrete for you -
17:45 - 17:48by sharing one
of my favorite applications of it. -
17:53 - 17:54I'll explain the LATE framework
-
17:54 - 17:57through a research question
that has fascinated me -
17:57 - 17:58for almost two decades.
-
17:59 - 18:00What is the causal effect
-
18:00 - 18:02of charter school attendance
on learning? -
18:02 - 18:04Charter schools are public schools
-
18:04 - 18:06that operate independently
-
18:06 - 18:09of traditional American
public school districts. -
18:09 - 18:12A charter, the right
to operate a public school -
18:12 - 18:14is typically awarded
for a limited period, -
18:15 - 18:18subject to renewal, conditional
on good school performance. -
18:18 - 18:20Charter schools
are free to structure -
18:20 - 18:22their curriculum
and school environment. -
18:22 - 18:24The most controversial difference
-
18:24 - 18:26between charters
and traditional public schools -
18:27 - 18:28is the fact that
the teachers and staff -
18:28 - 18:30who work at charter schools
-
18:30 - 18:32rarely belonged to labor unions.
-
18:32 - 18:35By contrast, most
big city public school teachers -
18:35 - 18:37work under union contracts.
-
18:38 - 18:41The 2010 documentary film
"Waiting for Superman" -
18:41 - 18:44features schools belonging to
the Knowledge is Power Program, -
18:44 - 18:45KIPP.
-
18:45 - 18:49KIPP schools are emblematic
of the high expectations, -
18:49 - 18:53sometimes also called "no excuses"
approach to public education. -
18:53 - 18:55The "no excuses" model features
-
18:55 - 18:58a long school day
and extended school year, -
18:58 - 18:59selective teacher hiring,
-
18:59 - 19:02and focuses on traditional
reading and math skills. -
19:03 - 19:06The American debate
over education reform -
19:06 - 19:08often focuses
on the achievement gap -- -
19:08 - 19:11that's shorthand
for large test score differences -
19:11 - 19:13by race and ethnicity.
-
19:13 - 19:15Because of its focus
on minority students, -
19:15 - 19:18KIPP is often central
in this debate -
19:18 - 19:19with supporters
pointing to the fact -
19:19 - 19:23that non-White KIPP students
have markedly higher test scores -
19:23 - 19:25than non-White students
from nearby schools. -
19:25 - 19:28KIPP skeptics, on the other hand,
-
19:28 - 19:30argue that KIPP's apparent success
-
19:30 - 19:32reflects the fact
that KIPP attracts families -
19:32 - 19:35whose children would be
more likely to succeed anyway. -
19:35 - 19:37Who's right?
-
19:37 - 19:39As you've probably guessed by now,
-
19:39 - 19:41a randomized trial
might prove decisive -
19:41 - 19:43in the debate
over schools like KIPP. -
19:44 - 19:45Like Nobel Prizes, though,
-
19:45 - 19:48seats at KIPP
are not randomly assigned. -
19:48 - 19:50Well, at least, not entirely.
-
19:51 - 19:52In fact,
-
19:52 - 19:55Massachusetts charter schools
with more applicants than seats -
19:55 - 19:57must offer their seats by lottery.
-
19:57 - 20:00Sounds like a good,
natural experiment. -
20:00 - 20:02A little over a decade ago,
-
20:02 - 20:04my collaborators and I
collected data -
20:04 - 20:06on KIPP admissions lotteries,
-
20:06 - 20:09laying the foundation
for two pioneering charter studies, -
20:09 - 20:12the first to use lotteries
to study KIPP. -
20:12 - 20:15Our KIPP analysis
is a classic IV story -
20:16 - 20:18because many students
offered a seat in the KIPP lottery -
20:19 - 20:20failed to show up in the fall,
-
20:20 - 20:24while a few not offered a seat
nevertheless find their way in. -
20:24 - 20:28This graphic shows KIPP
middle school applicants math scores -
20:28 - 20:30one year after applying to KIPP.
-
20:30 - 20:32The entries above the line
-
20:32 - 20:34show that KIPP applicants
who were offered a seat -
20:34 - 20:37have standardized
math scores close to zero -- -
20:37 - 20:39that is, near the state average.
-
20:39 - 20:42As before, we're working
with standardized score data -
20:42 - 20:45that has a mean of 0
and a standard deviation of 1. -
20:46 - 20:48Because KIPP applicants
start with 4th grade scores -
20:48 - 20:51that are roughly 0.3
standard deviations -
20:51 - 20:53below the state mean,
-
20:53 - 20:56achievement at the level
of the state average is impressive. -
20:57 - 21:01By contrast, the average math score
among those not offered a seat -
21:01 - 21:03is about -0.36 sigma --
-
21:03 - 21:07that is, 0.36 standard deviations
below the state mean, -
21:07 - 21:10a result typical for urban students
in Massachusetts. -
21:11 - 21:13Since lottery offers
are randomly assigned, -
21:13 - 21:17we could say with confidence
that the offer of a seat at KIPP -
21:17 - 21:20boost math scores
by an average of 0.36 sigma -- -
21:20 - 21:24a large effect
that's also statistically precise. -
21:24 - 21:26We can be confident
this isn't a chance finding. -
21:27 - 21:30What does an offer effect
of 0.36 sigma -
21:30 - 21:33tell us about the effects
of actually going to KIPP? -
21:34 - 21:36IV methods convert
KIPP offer effects -
21:36 - 21:38into KIPP attendance effects.
-
21:38 - 21:40I'll use this brief clip
-
21:40 - 21:43from my Marginal Revolution
University short course -
21:43 - 21:45to quickly review
the key assumptions -
21:45 - 21:46behind this conversion.
-
21:47 - 21:49- [Narrator] IV describes
a chain reaction. -
21:50 - 21:52Why do offers affect achievement?
-
21:52 - 21:55Probably because they affect
charter attendance, -
21:55 - 21:58and charter attendance
improves math scores. -
21:58 - 22:03The first link in the chain
called the First Stage -
22:03 - 22:06is the effect of the lottery
on charter attendance. -
22:06 - 22:08The Second Stage is the link
-
22:08 - 22:12between attending a charter
and an outcome variable -- -
22:12 - 22:14in this case, math scores.
-
22:14 - 22:18The instrumental variable,
or instrument for short, -
22:18 - 22:22is the variable that initiates
the chain reaction. -
22:23 - 22:26The effect of the instrument
on the outcome -
22:26 - 22:28is called the Reduced Form.
-
22:30 - 22:33This chain reaction can be
represented mathematically. -
22:34 - 22:38We multiply the First Stage --
the effect of winning on attendance, -
22:38 - 22:42by the Second Stage --
the effect of attendance on scores, -
22:42 - 22:44and we get the Reduced Form --
-
22:44 - 22:47the effect of winning
the lottery on scores. -
22:48 - 22:53The Reduced Form and First Stage
are observable and easy to compute. -
22:54 - 22:57However, the effect
of attendance on achievement -
22:57 - 22:59is not directly observed.
-
22:59 - 23:02This is the causal effect
we're trying to determine. -
23:03 - 23:06Given some important assumptions
we'll discuss shortly, -
23:06 - 23:08we can find the effect
of KIPP attendance -
23:08 - 23:11by dividing the Reduced Form
by the First Stage. -
23:13 - 23:15- [Joshua] IV eliminates
selection bias, -
23:15 - 23:17but like all of our tools,
-
23:17 - 23:19the solution builds
on a set of assumptions -
23:19 - 23:21not to be taken for granted.
-
23:22 - 23:25First, there must be
a substantial first stage -- -
23:25 - 23:27that is, the instrumental variable,
-
23:27 - 23:29winning or losing the lottery,
-
23:29 - 23:33must really change the variable
whose effect we're interested in -- -
23:33 - 23:34here, KIPP attendance.
-
23:35 - 23:38In this case, the first stage
is not really in doubt. -
23:38 - 23:39Winning the lottery
-
23:39 - 23:42makes KIPP attendance
much more likely. -
23:42 - 23:44Not all IV stories are like that.
-
23:45 - 23:48Second, the instrument must be
as good as randomly assigned, -
23:48 - 23:52meaning lottery winners and losers
have similar characteristics. -
23:52 - 23:55This is the independence assumption.
-
23:55 - 23:59Of course, KIPP lottery wins
really are randomly assigned. -
23:59 - 24:02Still, we should check
for balance and confirm -
24:02 - 24:03that winners and losers
-
24:03 - 24:07have similar family backgrounds,
similar aptitudes, and so on. -
24:07 - 24:10In essence, we're checking
to ensure KIPP lotteries are fair, -
24:11 - 24:14with no group of applicants
suspiciously likely to win. -
24:15 - 24:18Finally, we require
the instrument change outcomes -
24:18 - 24:20solely through
the variable of interest -- -
24:20 - 24:21in this case, attending KIPP.
-
24:22 - 24:25This assumption is called
the Exclusion Restriction. -
24:27 - 24:29The causal effect
of KIPP attendance -
24:29 - 24:30can therefore be written
-
24:30 - 24:33as the ratio of the effect
of offers on scores -
24:33 - 24:34in the numerator
-
24:34 - 24:36over the effect of offers
on KIPP enrollment -
24:36 - 24:37in the denominator.
-
24:37 - 24:40The numerator in this IV formula --
-
24:40 - 24:43that is, the direct effect
of the instrument on outcomes -
24:43 - 24:44has a special name.
-
24:44 - 24:47This is called the Reduced Form.
-
24:47 - 24:49The denominator is the First Stage.
-
24:49 - 24:52The exclusion restriction
is often the trickiest -
24:52 - 24:55or most controversial part
of an IV story. -
24:55 - 24:58Here, the exclusion restriction
amounts to the claim -
24:58 - 25:02that the 0.36 score differential
between lottery winners and losers -
25:02 - 25:04is entirely attributable
-
25:04 - 25:08to the 0.74 win/loss difference
in attendance rates. -
25:08 - 25:09Plugging in the numbers,
-
25:09 - 25:13the effect of KIPP attendance
works out to be 0.48 sigma, -
25:13 - 25:15almost half
a standard deviation gain -
25:15 - 25:17in math scores --
-
25:17 - 25:19that's a remarkably large effect.
-
25:19 - 25:24Who exactly benefits
so spectacularly from KIPP? -
25:24 - 25:27Does everyone who applies
to KIPP see such large gains? -
25:28 - 25:29LATE answers this question.
-
25:30 - 25:33The LATE interpretation
of the KIPP IV empirical strategy -
25:33 - 25:37is illuminated
by the biblical story of Passover, -
25:37 - 25:40which explains that there are
four types of children, -
25:40 - 25:42each with characteristic behaviors.
-
25:42 - 25:45To keep track of these children
and their behavior, -
25:45 - 25:47I'll give them alliterative names.
-
25:47 - 25:51Applicants like Alvaro
are dying to go to KIPP. -
25:51 - 25:53If Alvaro loses the KIPP lottery,
-
25:53 - 25:56his mother finds a way
to enroll him in KIPP anyway, -
25:56 - 25:58perhaps by reapplying.
-
25:58 - 26:01Applicants like Camila
are happy to go to KIPP -
26:01 - 26:03if they win a seat in the lottery,
-
26:03 - 26:06but stoically accept
the verdict, if they lose. -
26:06 - 26:09Finally, applicants like Normando
-
26:09 - 26:12worry about long days
and lots of homework at KIPP. -
26:12 - 26:14Normando doesn't really want to go
-
26:14 - 26:17and refuses to go to KIPP
when told that he won the lottery. -
26:18 - 26:20Normando was called a never-taker
-
26:20 - 26:22because win or lose,
he doesn't go to KIPP. -
26:22 - 26:24At the other end
of KIPP commitment, -
26:24 - 26:27Alvaro is called an always-taker.
-
26:27 - 26:29He'll happily take a seat
when offered -
26:29 - 26:33while his mother simply finds a way
to make it happen for him, -
26:33 - 26:34even when he loses.
-
26:34 - 26:36For Alvaro and Normando both,
-
26:36 - 26:41choice of school, KIPP, traditional,
is unaffected by the lottery. -
26:41 - 26:45Camila is the type of applicant
who gives IV its power. -
26:45 - 26:48The instrument determines
her treatment status. -
26:48 - 26:52IV strategies depend
on applicants like Camilla, -
26:52 - 26:54who are called compliers.
-
26:54 - 26:57This term comes from the world
of randomized trials -
26:57 - 26:58introduced earlier.
-
26:58 - 27:00As we've already discussed,
-
27:00 - 27:05many randomized trials randomize
only the opportunity to be treated -
27:05 - 27:08while the decision
to comply with the treatment -
27:08 - 27:10remains voluntary and non-random.
-
27:11 - 27:13RCT compliers are those
who take treatment -
27:13 - 27:15when the offer of treatment is made,
-
27:15 - 27:17but not otherwise.
-
27:17 - 27:18With lottery instruments,
-
27:18 - 27:21LATE is the effect
of KIPP attendance on Camila -
27:21 - 27:23and other compliers like her
-
27:23 - 27:26who enroll at KIPP, take treatment
-
27:26 - 27:28when offered treatment
through the lottery, -
27:28 - 27:29but not otherwise.
-
27:30 - 27:31IV methods are uninformative
-
27:31 - 27:35for always-takers like Alvaro
and never-takers like Normando -
27:35 - 27:37because the instrument
is unrelated -
27:37 - 27:39to their treatment status.
-
27:39 - 27:42Hey, didn't I say
there are four types of children? -
27:42 - 27:46A fourth type of child in IV theory
behaves perversely. -
27:46 - 27:48Every family has one!
-
27:48 - 27:51These defiant children
enroll in KIPP -
27:51 - 27:52only when they lose the lottery.
-
27:53 - 27:54Actually, the LATE theorem
-
27:54 - 27:57requires us to assume
there are few defiers -- -
27:57 - 27:59that seems like
a reasonable assumption -
27:59 - 28:00for charter lottery instruments,
-
28:00 - 28:02if not in life.
-
28:02 - 28:04The LATE theorem is sometimes seen
-
28:04 - 28:07as limiting the relevance
of econometric estimates -
28:07 - 28:10because it focuses attention
on groups of compliers. -
28:11 - 28:12Yet, the population of compliers
-
28:12 - 28:15is a group we'd very much like
to learn about. -
28:15 - 28:16In the KIPP example,
-
28:16 - 28:20compliers are children
likely to be drawn into KIPP -
28:20 - 28:21were the school to expand
-
28:21 - 28:24and offer additional seats
in a lottery. -
28:24 - 28:26How relevant is this?
-
28:26 - 28:27A few years ago,
-
28:27 - 28:31Massachusetts indeed allowed
thriving charter schools to expand. -
28:31 - 28:34A recent study
by some of my lab mates -
28:34 - 28:35shows that LATE estimates,
-
28:35 - 28:37like the one
we just computed for KIPP, -
28:37 - 28:38predict learning gains
-
28:38 - 28:41at the schools created
by charter expansion. -
28:46 - 28:47LATE isn't just a theorem --
-
28:47 - 28:49it's a framework.
-
28:49 - 28:53The LATE framework can be used
to estimate the entire distribution -
28:53 - 28:55of potential outcomes for compliers
-
28:55 - 28:59as if we really did have
a randomized trial for this group. -
28:59 - 29:01Although the theory
behind this fact -
29:01 - 29:03is necessarily technical,
-
29:03 - 29:07the value of the framework
is easily appreciated in practice. -
29:07 - 29:10By way of illustration,
recall that the KIPP study -
29:10 - 29:14is motivated in part by differences
in test scores by race. -
29:14 - 29:17Let's look at the distribution
of 4th grade scores, -
29:17 - 29:19separately by race,
-
29:19 - 29:22for applicants to Boston
charter middle schools. -
29:22 - 29:24The two sides of this figure
-
29:24 - 29:28show distributions for treated
and untreated compliers. -
29:28 - 29:32Treated compliers are compliers
offered a charter seat in a lottery, -
29:32 - 29:35while untreated compliers
are not offered a seat. -
29:35 - 29:37Because these are 4th grade scores,
-
29:37 - 29:40while middle school begins
in 5th or 6th grade, -
29:40 - 29:42the two sides
of the figure are similar. -
29:42 - 29:46Both sides show score distributions
for Black applicants -
29:46 - 29:47shifted to the left
-
29:47 - 29:50of the corresponding
score distributions for Whites. -
29:50 - 29:52By 8th grade,
-
29:52 - 29:56treated compliers have completed
middle school at a Boston charter, -
29:56 - 29:58while untreated compliers
have remained -
29:58 - 30:00in traditional public school.
-
30:00 - 30:02Remarkably, this next graphic
-
30:02 - 30:05shows that the 8th grade
score distributions -
30:05 - 30:07of Black and White treated compliers
-
30:07 - 30:08are indistinguishable.
-
30:09 - 30:12Boston charter middle schools
closed the achievement gap. -
30:13 - 30:14But for the untreated,
-
30:14 - 30:17Black and White score distributions
remained distinct -
30:17 - 30:20with Black students
behind White students -
30:20 - 30:22as they were in 4th grade.
-
30:22 - 30:24Boston charters closed
the achievement gap -
30:24 - 30:26because those who enter
charter schools -
30:26 - 30:27the farthest behind
-
30:27 - 30:30tend to gain the most
from charter enrollment. -
30:31 - 30:34I elaborate on this point
in the print version of this talk. -
30:39 - 30:40Remember the puzzle
-
30:40 - 30:43of negative Chicago
exam school effects? -
30:43 - 30:47I'll finish the scientific
part of my talk by using IV and RD -
30:47 - 30:50to explain this surprising finding.
-
30:50 - 30:53The resolution of this puzzle
starts with the fact -
30:53 - 30:56that economic reasoning
is about alternatives. -
30:57 - 31:00So what's the alternative
to an exam school education? -
31:00 - 31:03For most applicants
to Chicago exam schools, -
31:03 - 31:07the leading non-exam alternative
is a traditional public school. -
31:08 - 31:11But many of Chicago's rejected
exam school applicants -
31:11 - 31:13enroll in a charter school.
-
31:14 - 31:15Exam school offers
-
31:15 - 31:18therefore reduce the likelihood
of charter school attendance. -
31:18 - 31:22Specifically, exam schools
divert applicants -
31:22 - 31:24away from high schools
-
31:24 - 31:26in the Noble Network
of Charter Schools. -
31:27 - 31:30Noble, with pedagogy
much like KIPP, -
31:30 - 31:33is one of Chicago's most visible
charter providers. -
31:33 - 31:37Also like KIPP, convincing evidence
on Noble effectiveness -
31:37 - 31:39comes from admissions lotteries.
-
31:39 - 31:42The x-axis in this graphic
-
31:42 - 31:45shows lottery offer effects
on years enrolled at Noble. -
31:46 - 31:48This is the Noble first stage,
-
31:48 - 31:51for an IV setup that uses a dummy
-
31:51 - 31:53indicating Noble lottery offers
-
31:53 - 31:56as an instrument
for Noble enrollment. -
31:56 - 31:59Now this graphic has a feature
that distinguishes it -
31:59 - 32:02from the simpler KIPP analysis.
-
32:02 - 32:04The plot shows first-stage effects
-
32:04 - 32:05for two groups:
-
32:05 - 32:07one for Noble applicants
-
32:07 - 32:11who live in Chicago's lowest income
neighborhoods, Tier 1, -
32:11 - 32:13and one for Noble applicants
-
32:13 - 32:15who live in higher-income areas,
-
32:15 - 32:16Tier 3.
-
32:16 - 32:18Remember the IV chain reaction?
-
32:19 - 32:20Each point in this graphic
-
32:20 - 32:24has coordinates given
by first-stage reduced form -
32:24 - 32:27and therefore implies
an IV estimate. -
32:27 - 32:30The effect of Noble enrollment
on ACT scores -
32:30 - 32:32is the ratio
of Reduced-Form coordinate -
32:32 - 32:34to First-Stage coordinate.
-
32:34 - 32:37The graphic shows two such ratios.
-
32:37 - 32:41The relevant results
for Tier 1 are 0.35, -
32:41 - 32:44while for Tier 3, we have 0.33 --
-
32:44 - 32:45not bad.
-
32:45 - 32:48For Noble applicants
from both tiers, -
32:48 - 32:50these First-Stage
and Reduced-Form estimates -
32:50 - 32:52imply a yearly
Noble enrollment effect -
32:52 - 32:55of about a third
of a standard deviation gain -
32:55 - 32:57in ACT math scores.
-
32:58 - 32:59Notice there's also a line
-
32:59 - 33:02connecting the two IV estimates
in the figure. -
33:02 - 33:05Because this line
passes through the origin, -
33:05 - 33:07its slope, "rise over run,"
-
33:07 - 33:10is about equal
to the two IV estimates -- -
33:10 - 33:13in this case,
the slope is about 0.34. -
33:14 - 33:17The fact that the line
passes through 0,0 -
33:17 - 33:19is significant for another reason.
-
33:19 - 33:23By this fact, we've substantiated
the exclusion restriction. -
33:23 - 33:26Specifically,
the exclusion restriction -
33:26 - 33:28says that given a group
-
33:28 - 33:31for which Noble offers
are unrelated to Noble enrollment, -
33:31 - 33:33we should expect to see
-
33:33 - 33:360 reduced-form effect
of these offers -
33:36 - 33:39made to applicants in that group.
-
33:39 - 33:43How consistent is the evidence
for a Noble cause learning gain -
33:43 - 33:46on the order
of 0.34 sigma per year? -
33:46 - 33:47In this next graphic,
-
33:47 - 33:50we've added 12 more points
to the original 2. -
33:50 - 33:53The red points here
show First-Stage and Reduced-Form -
33:53 - 33:57Noble offer effects
for 12 additional groups, -
33:57 - 33:592 more tiers and 12 groups
-
33:59 - 34:02defined by demographic
characteristics -
34:02 - 34:05related to race, sex,
family income, -
34:05 - 34:06and baseline scores.
-
34:06 - 34:08Although not a perfect fit,
-
34:08 - 34:12these points cluster around a line
with slope 0.36 sigma -
34:12 - 34:16much like the line we saw earlier
for applicants from Tiers 1 and 3. -
34:17 - 34:18You're likely now wondering
-
34:18 - 34:21what the Noble IV
estimates in this figure -
34:21 - 34:24have to do
with exam school enrollment. -
34:24 - 34:25Here's the answer.
-
34:26 - 34:30The blue line in this new graphic
shows, as we should expect, -
34:30 - 34:32that exam school exposure jumps up
-
34:32 - 34:35for applicants who clear
their qualifying cutoff. -
34:35 - 34:37At the same time,
-
34:37 - 34:39the red line shows
that Noble school enrollment -
34:39 - 34:42clearly falls at the same point.
-
34:42 - 34:46This is the diversion effect
of exam school offers -
34:46 - 34:48on Noble enrollment.
-
34:48 - 34:51Many kids offered
an exam school seat -
34:51 - 34:55prefer that exam school seat
to enrollment at Noble. -
34:55 - 34:58IV affords us the opportunity
to go out on a limb -
34:58 - 35:00with strong claims
about the mechanism -
35:00 - 35:02behind the causal effect.
-
35:02 - 35:03Here's a strong causal claim
-
35:03 - 35:07regarding why Chicago exam schools
reduce achievement. -
35:08 - 35:09The primary force
-
35:09 - 35:12driving reduced-form
exam school qualification effects -
35:12 - 35:15on ACT scores, I claim,
-
35:15 - 35:19is the effect of exam school offers
on Noble enrollment. -
35:19 - 35:21In support of this claim,
-
35:21 - 35:23consider the points
plotted here in blue, -
35:23 - 35:27all well to the left
of 0 on the x-axis. -
35:27 - 35:29These points are negative
-
35:29 - 35:32because they mark the effect
of exam school qualification -
35:32 - 35:34on Noble school enrollment
-
35:34 - 35:36for particular groups of applicants.
-
35:37 - 35:38Now we've already seen
-
35:38 - 35:41that Noble applicants
offered a Noble seat -
35:41 - 35:44realize large ACT
math gains as a result. -
35:45 - 35:48Now consider exam school offers
-
35:48 - 35:50as an instrument
for Noble enrollment. -
35:51 - 35:53As always, IV is a chain reaction.
-
35:54 - 35:57If exam school qualification
reduces time at Noble -
35:57 - 35:58by 0.37 years,
-
35:58 - 36:00and each year of Noble enrollment
-
36:00 - 36:04boosts ACT math scores
by about 0.36 sigma, -
36:04 - 36:06we should expect
reduced-form effects -
36:06 - 36:08of exam school qualification
-
36:08 - 36:10to reduce ACT scores
-
36:10 - 36:12by the product
of these two numbers -- -
36:12 - 36:15that is, by about 0.13 sigma.
-
36:15 - 36:17The reduced-form
qualification effects -
36:17 - 36:18at the left of the figure
-
36:18 - 36:20are broadly consistent with this.
-
36:21 - 36:25They cluster closer to -0.16
than to -0.13, -
36:25 - 36:28but that difference is well
within the sampling variance -
36:28 - 36:29of the underlying estimates.
-
36:30 - 36:31The causal story told here
-
36:32 - 36:35postulates diversion
away from charter schools -
36:35 - 36:38as the mechanism
by which exam school offers -
36:38 - 36:39affect achievement.
-
36:40 - 36:42In other words,
it's Noble enrollment -
36:42 - 36:45that's presumed to satisfy
an exclusion restriction -
36:45 - 36:47when we use exam school offers
-
36:47 - 36:49as an instrumental variable.
-
36:49 - 36:51Importantly, as we saw before,
-
36:51 - 36:55the line in this final graphic,
with two sets of 14 points, -
36:55 - 36:57runs through the origin.
-
36:57 - 37:00This fact supports
our new exclusion restriction. -
37:01 - 37:02For any applicant group
-
37:02 - 37:03for which exam school offers
-
37:03 - 37:06have little or no effect
on Noble school enrollment, -
37:06 - 37:10we should also see
ACT scores unchanged. -
37:10 - 37:11At the same time,
-
37:11 - 37:15because the blue and red dots
cluster around the same line, -
37:15 - 37:17the IV estimates of Noble school
enrollment effects -
37:17 - 37:21generated by both Noble
and exam school offers -
37:21 - 37:23are about the same.
-
37:23 - 37:25I hope this empirical story
-
37:25 - 37:28convinces you
of the power of IV and RD -
37:28 - 37:30to generate new causal knowledge.
-
37:30 - 37:32For decades,
I've been lucky to work -
37:32 - 37:35on many equally engaging
empirical problems. -
37:40 - 37:43I computed the draft
lottery IV estimates -
37:43 - 37:45in my Princeton Ph.D. thesis
-
37:45 - 37:47on a big, hairy mainframe monster,
-
37:47 - 37:50using 9-track tapes and leased space
-
37:50 - 37:51on a communal hard drive.
-
37:52 - 37:53Princeton graduate students
-
37:53 - 37:55learned to mount
and manipulate tape reels -
37:55 - 37:57the size of a cheesecake.
-
37:58 - 38:02Thankfully, empirical work today
is a little less labor-intensive. -
38:02 - 38:05What else has improved
in the modern empirical era? -
38:06 - 38:09in a 2010 article, Steve Pischke
and I coined the phrase -
38:09 - 38:11"Credibility Revolution."
-
38:11 - 38:13By this, we mean economic shift
-
38:13 - 38:16towards transparent
empirical strategies -
38:16 - 38:18applied to concrete
causal questions, -
38:19 - 38:22like the questions David Card
has studied so convincingly. -
38:23 - 38:25The econometrics of my school days
-
38:25 - 38:28focused more on models
than on questions. -
38:28 - 38:31The modeling concerns
of that era have mostly faded, -
38:31 - 38:35but econometricians have since
found much to contribute. -
38:35 - 38:37I'll save my personal lists
-
38:37 - 38:39of greatest hits
and exciting new artists -
38:39 - 38:41for the print version of this lecture.
-
38:41 - 38:43I'll wrap up here by saying
-
38:43 - 38:46that I'm proud to be part
of the contemporary -
38:46 - 38:48empirical economics enterprise,
-
38:48 - 38:51and I'm gratified beyond words
-
38:51 - 38:54to have been recognized
for contributing to it. -
38:54 - 38:56Back at Princeton, in the late '80s,
-
38:56 - 38:58my graduate school classmates
and I chuckled -
38:58 - 39:00reading Ed Leamer's lament
-
39:00 - 39:04that no economist takes another
economist's empirical work seriously. -
39:05 - 39:07This is no longer true.
-
39:07 - 39:11Empirical work today aspires
to tell convincing causal stories. -
39:11 - 39:13Not that every effort succeeds --
-
39:13 - 39:15far from it.
-
39:15 - 39:18But as any economics job
market candidate will tell you, -
39:18 - 39:22empirical work carefully executed
and clearly explained -
39:22 - 39:23is taken seriously indeed --
-
39:24 - 39:27that is a measure
of our enterprise's success. -
39:27 - 39:30♪ [music] ♪
-
39:34 - 39:36- [Narrator] If you'd like
to learn more from Josh, -
39:36 - 39:39check out his free course
"Mastering Econometrics." -
39:39 - 39:41If you'd like to explore
Josh's research, -
39:41 - 39:42check out the links
in the description, -
39:42 - 39:45or you can click to watch more
of Josh's videos. -
39:46 - 39:48♪ [music] ♪
- Title:
- Joshua Angrist Nobel Prize Lecture 2021
- ASR Confidence:
- 0.85
- Description:
-
Joshua Angrist, winner of The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel (2021), delivers his Nobel Prize lecture on "Empirical Strategies in Economics: Illuminating the Path from Cause to Effect".
**LEARN MORE ABOUT JOSH ANGRIST’S RESEARCH**
Blueprint Labs at MIT: https://blueprintlabs.mit.edu/
Josh Angrist’s Research Publications: https://economics.mit.edu/faculty/angrist/publications
Josh Angrist’s Working Papers: https://economics.mit.edu/faculty/angrist/papers**LEARN MORE ABOUT JOSH ANGRIST’S BOOKS**
Mostly Harmless Econometrics!: http://www.mostlyharmlesseconometrics.com/
Mastering ‘Metrics: http://www.masteringmetrics.com/**INSTRUCTOR RESOURCES**
Mastering Econometrics course: https://mru.io/nk7
Econometrics test bank: https://mru.io/of6
More professor resources: https://mru.io/9gq**MORE LEARNING**
Mastering Econometrics course: https://mru.io/nk7
Receive updates when we release new videos: https://mru.io/ri7
More from Marginal Revolution University: https://mru.io/0nf - Video Language:
- English
- Team:
- Marginal Revolution University
- Duration:
- 39:55
LindsayMRU edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
LindsayMRU edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
LindsayMRU edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
Kirstin Cosper edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
Kirstin Cosper edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
Kirstin Cosper edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
Kirstin Cosper edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 | ||
Kirstin Cosper edited English subtitles for Joshua Angrist Nobel Prize Lecture 2021 |