WEBVTT
00:00:00.900 --> 00:00:02.760
♪ [music] ♪
00:00:03.800 --> 00:00:05.600
- [Narrator] Welcome to
Nobel conversations.
00:00:07.250 --> 00:00:10.240
In this episode, Josh Angrist
and Guido Imbens,
00:00:10.240 --> 00:00:11.920
sit down with Isaiah Andrews
00:00:11.920 --> 00:00:14.303
to discuss how their research
was initially received
00:00:14.900 --> 00:00:17.090
and how they responded
to criticism.
00:00:18.700 --> 00:00:21.400
At the time, did you feel like
you are on to something,
00:00:21.680 --> 00:00:25.152
you felt this was the beginning
of a whole line of work
00:00:25.152 --> 00:00:27.202
that you felt like was going
to be important or...?
00:00:27.600 --> 00:00:30.000
Not so much that it was
a whole line of work,
00:00:30.000 --> 00:00:31.894
but certainly I felt like,
"Wow, this--"
00:00:32.277 --> 00:00:35.045
- [Josh] We proved something
that people didn't know before,
00:00:35.045 --> 00:00:36.340
that it was worth knowing.
00:00:36.340 --> 00:00:41.143
Yeah, going back compared to
my job market papers having--
00:00:41.600 --> 00:00:45.560
I felt this was actually
a very clear crisp result.
00:00:46.400 --> 00:00:49.530
But there were definitely
was mixed reception
00:00:49.530 --> 00:00:52.420
and I don't think anybody
said that,
00:00:52.420 --> 00:00:55.461
"Oh, wow, this is already,
something."
00:00:55.461 --> 00:00:59.386
No, which is the nightmare scenario
for a researcher
00:01:00.230 --> 00:01:02.003
where you think you've
discovered something
00:01:02.003 --> 00:01:04.461
and then somebody else says,
"Oh, I knew that."
00:01:05.000 --> 00:01:07.220
But there were definitely was
a need to convince people
00:01:07.220 --> 00:01:10.370
that this was worth knowing,
that instrumental variables
00:01:10.370 --> 00:01:12.687
estimates a causal effect
for compliers.
00:01:13.200 --> 00:01:16.178
Yeah, but even though it
took a long time
00:01:16.178 --> 00:01:19.348
to convince a bigger audience,
00:01:19.820 --> 00:01:24.346
sometimes even fairly quickly,
the reception was pretty good
00:01:24.800 --> 00:01:26.645
among a small group of people.
00:01:27.200 --> 00:01:31.297
Gary, clearly liked it a lot
from the beginning
00:01:31.800 --> 00:01:35.645
and I remember, because at that point
Josh had left for Israel,
00:01:35.645 --> 00:01:38.886
but I remember explaining it
to Don Ruben
00:01:39.696 --> 00:01:43.700
and he was like,
"Yeah, this really is something here."
00:01:43.700 --> 00:01:47.100
Not right away though,
Don took some convincing.
00:01:47.500 --> 00:01:49.150
By the time you got to Don,
00:01:49.150 --> 00:01:51.226
there have been some back
and forth with him
00:01:51.226 --> 00:01:53.304
and in correspondence actually.
00:01:53.700 --> 00:01:57.103
But I remember at some point
getting a call or email from him
00:01:57.103 --> 00:02:00.020
saying that he was sitting
at the airport in Rome
00:02:00.020 --> 00:02:03.700
and looking at the paper
and thinking,
00:02:03.700 --> 00:02:07.000
"Yeah, no actually,
you guys are onto something."
00:02:07.490 --> 00:02:08.594
We were happy about that
00:02:08.594 --> 00:02:10.550
but that took longer
than I think you remember.
00:02:11.030 --> 00:02:12.500
It wasn't right away
00:02:12.500 --> 00:02:13.700
[laughter]
00:02:13.700 --> 00:02:16.230
because I know that I was back
in Israel by the time that happened.
00:02:16.500 --> 00:02:18.750
I'd left for Israel
in the summer of--
00:02:19.390 --> 00:02:21.190
I was only at Harvard
for two years.
00:02:21.190 --> 00:02:22.540
We had that one year.
00:02:22.540 --> 00:02:25.700
It is remarkable, I mean, that
one year was so fateful for us.
00:02:25.900 --> 00:02:27.200
- [Guido] Yes.
00:02:27.690 --> 00:02:30.200
I think we understood there was
something good happening,
00:02:30.200 --> 00:02:33.700
but maybe we didn't think it was
life-changing, only in retrospect.
00:02:33.700 --> 00:02:35.620
♪ [music] ♪
00:02:35.620 --> 00:02:37.495
- [Isaiah] As you said, it sounds like
a small group of people
00:02:37.495 --> 00:02:39.190
were initially quite receptive,
00:02:39.190 --> 00:02:42.190
perhaps took some time for
a broader group of people
00:02:43.090 --> 00:02:45.912
to come around to seeing
the LATE framework
00:02:45.912 --> 00:02:47.620
as a valuable way to look
at the world.
00:02:47.620 --> 00:02:50.100
I guess, in over the course of that,
00:02:50.100 --> 00:02:52.128
were their periods
where you thought,
00:02:52.128 --> 00:02:54.450
maybe the people saying
this wasn't a useful way
00:02:54.450 --> 00:02:55.751
to look at the world were right?
00:02:55.751 --> 00:02:58.260
Did you get discouraged?
How did you think about?
00:02:58.400 --> 00:03:00.900
I don't think I was discouraged
but the people who were saying
00:03:00.900 --> 00:03:05.900
that we're smart people,
well informed econometricians,
00:03:05.900 --> 00:03:07.800
sophisticated readers
00:03:08.900 --> 00:03:11.800
and I think the substance
of the comment was,
00:03:11.800 --> 00:03:14.297
this is not what econometrics
is about.
00:03:14.297 --> 00:03:20.572
Econometrics was being transmitted
at that time was about structure.
00:03:21.300 --> 00:03:24.490
There was this idea that
there's structure in the economy
00:03:25.100 --> 00:03:27.200
and it's our job to discover it
00:03:27.200 --> 00:03:30.952
and what makes it structure
is it's essentially invariant
00:03:32.570 --> 00:03:34.900
and so we're saying,
in the LATE theorem,
00:03:34.900 --> 00:03:37.699
that every instrument produces
its own causal effect,
00:03:37.699 --> 00:03:41.386
which is in contradiction to that
to some extent
00:03:41.386 --> 00:03:43.802
and so that was
where the tension was.
00:03:43.802 --> 00:03:45.551
People didn't want
to give up that idea.
00:03:46.300 --> 00:03:50.369
Yeah, I remember once
people were started
00:03:51.200 --> 00:03:55.664
arguing more vocally against that,
00:03:56.900 --> 00:03:59.483
that never really
bothered me that much.
00:03:59.483 --> 00:04:03.051
It seemed clear that
we had a result there
00:04:03.051 --> 00:04:05.878
and it became somewhat
controversial,
00:04:05.878 --> 00:04:08.395
but controversial in a good way.
00:04:08.620 --> 00:04:10.190
It was clear that people felt
00:04:10.820 --> 00:04:13.835
they had to come out against it
because--
00:04:13.970 --> 00:04:15.649
Well, I think what
we think it's good now
00:04:17.426 --> 00:04:19.238
we might not have loved it
at the time.
00:04:20.168 --> 00:04:22.984
I remember being somewhat,
the more upset--
00:04:22.984 --> 00:04:24.780
there was some dinner
where someone said,
00:04:24.780 --> 00:04:27.455
"No, no, no, that paper with Josh,
00:04:28.855 --> 00:04:30.749
that was doing a disservice
to the profession."
00:04:32.050 --> 00:04:33.850
We definitely had
reactions like that.
00:04:35.410 --> 00:04:38.200
At some level, that may be
indicative of the culture
00:04:38.400 --> 00:04:40.000
in general in economics
at the time.
00:04:41.400 --> 00:04:44.097
I thought back later,
what if that'd happened now,
00:04:44.600 --> 00:04:47.682
if I was a senior person sitting
in that conversation,
00:04:48.200 --> 00:04:51.898
I would call that out because it
really was not appropriate--
00:04:53.000 --> 00:04:54.200
- [Josh] It wasn't so bad.
00:04:54.600 --> 00:04:56.600
I think the criticism is--
00:04:57.700 --> 00:04:59.298
It wasn't completely misguided,
00:05:00.070 --> 00:05:01.351
it was maybe wrong.
00:05:01.800 --> 00:05:04.485
No, no, but you can say
the paper is wrong
00:05:05.280 --> 00:05:06.440
but it's saying that
00:05:06.440 --> 00:05:08.128
it's a disservice
to the profession,
00:05:08.128 --> 00:05:09.300
that's not really--
00:05:09.300 --> 00:05:10.300
Personal.
00:05:10.300 --> 00:05:12.646
Yes, and doing that, not to me,
00:05:12.646 --> 00:05:14.442
but in front of
my senior colleagues.
00:05:15.191 --> 00:05:17.369
But nobody was saying
the result was wrong
00:05:17.369 --> 00:05:18.700
and I remember also,
00:05:18.700 --> 00:05:21.579
some of the comments
were thought-provoking
00:05:21.579 --> 00:05:23.059
so we had some negative reviews,
00:05:23.059 --> 00:05:25.861
I think on the average
causal response paper.
00:05:26.500 --> 00:05:30.361
Somebody said, "These compliers
you can't figure out who they are."
00:05:31.450 --> 00:05:33.220
Right, it's one thing to say
00:05:33.220 --> 00:05:35.040
you're estimating
the effect of treatment
00:05:35.040 --> 00:05:36.788
on the treated
or something like that.
00:05:36.788 --> 00:05:38.400
You can tell me who's treated,
00:05:38.700 --> 00:05:42.289
people in the CPS,
you can't tell me who's a complier.
00:05:42.929 --> 00:05:44.679
So that was a legitimate challenge.
00:05:44.679 --> 00:05:47.800
That's certainly fair
and I can see why
00:05:49.880 --> 00:05:53.502
that part made people
a little uneasy and uncomfortable.
00:05:54.300 --> 00:05:56.400
But at the same time,
00:05:56.900 --> 00:06:00.244
because it showed that you couldn't
really go beyond that,
00:06:00.800 --> 00:06:03.775
it was very useful thing
to realize.
00:06:04.630 --> 00:06:06.180
I remember on the day,
00:06:06.500 --> 00:06:09.771
we got to the key result
that I was thinking,
00:06:09.771 --> 00:06:13.113
"Wow, this is as good as it gets.
00:06:14.221 --> 00:06:16.978
Here we actually have an insight
but clearly--"
00:06:17.500 --> 00:06:19.250
And we had to sell it
at some point.
00:06:19.480 --> 00:06:21.261
For quite a few years,
we had to sell it
00:06:23.480 --> 00:06:24.892
and it's proven to be quite useful.
00:06:25.500 --> 00:06:28.761
I don't think we understood that
it would be so useful at the time.
00:06:28.761 --> 00:06:29.871
No.
00:06:30.170 --> 00:06:34.600
I did feel early on this was
a substantial insight.
00:06:34.600 --> 00:06:36.440
- [Josh] Yeah we [did] something.
00:06:36.440 --> 00:06:40.041
But I did not think
goals were there.
00:06:40.700 --> 00:06:42.600
I don't think we were aiming
for the Nobel.
00:06:42.600 --> 00:06:43.730
[laughter]
00:06:43.730 --> 00:06:46.243
We were very happy to get
that note in Econometrica.
00:06:46.859 --> 00:06:48.829
♪ [music] ♪
00:06:49.770 --> 00:06:51.500
- [Isaiah] Are there factors
or are ways of approaching problems
00:06:51.500 --> 00:06:54.186
that lead people to be better
at recognizing the good stuff
00:06:54.186 --> 00:06:56.600
and taking the time to do it
as opposed to dismissing it?
00:06:56.600 --> 00:06:57.830
- [Josh] Sometimes
I think it's helpful.
00:06:57.830 --> 00:06:59.478
If you're trying to
convince somebody
00:06:59.478 --> 00:07:01.247
that you have something
useful to say
00:07:01.900 --> 00:07:04.176
and maybe they don't
speak your language,
00:07:04.894 --> 00:07:06.541
you might need
to learn their language.
00:07:06.761 --> 00:07:07.910
Yes, yes, exactly.
00:07:07.910 --> 00:07:11.736
That's what we did with Don,
we figured out how to--
00:07:11.736 --> 00:07:14.052
I remember we had a very hard time
00:07:14.052 --> 00:07:15.816
explaining the exclusion restriction
to Don,
00:07:17.430 --> 00:07:18.993
maybe rightfully so,
00:07:19.804 --> 00:07:21.948
I think Guido and I
eventually figured out
00:07:21.948 --> 00:07:24.420
that it wasn't formulated
very clearly,
00:07:25.400 --> 00:07:27.450
and we came up
with a way to do that
00:07:27.450 --> 00:07:29.316
in the potential outcomes framework
00:07:29.316 --> 00:07:32.218
that I think worked
for the three of us.
00:07:32.218 --> 00:07:33.419
- [Guido] Yeah.
00:07:33.419 --> 00:07:35.454
Well, it worked for
the bigger literature
00:07:35.454 --> 00:07:37.639
but I think what you're saying
there is exactly right,
00:07:37.639 --> 00:07:40.860
you need to figure out
how not just say,
00:07:40.860 --> 00:07:43.894
"Okay well, I've got this language
and this this works great
00:07:43.894 --> 00:07:45.900
and I've got to convince someone
else to use the language.
00:07:45.900 --> 00:07:48.188
You could first figure out
what language they're using
00:07:48.680 --> 00:07:51.028
and then only then,
can you try to say,
00:07:51.028 --> 00:07:53.140
"Well, but here you thinking of it
this way,"
00:07:53.140 --> 00:07:56.880
but that's actually
a pretty hard thing to do,
00:07:56.880 --> 00:07:59.098
you get someone from
a different discipline,
00:07:59.098 --> 00:08:02.300
convincing them, two junior faculty
in a different department
00:08:02.300 --> 00:08:04.366
actually have something
to say to you,
00:08:04.596 --> 00:08:06.516
that takes a fair amount of effort.
00:08:07.500 --> 00:08:09.782
Yeah, I wrote Don
a number of times,
00:08:10.420 --> 00:08:11.868
in fairly long letters.
00:08:11.868 --> 00:08:13.805
I remember thinking
this is worth doing,
00:08:14.600 --> 00:08:16.006
that if I could convince Don
00:08:16.780 --> 00:08:19.444
that would validate the framework
to some extent.
00:08:20.300 --> 00:08:22.924
I think both you and Don were
00:08:22.924 --> 00:08:25.000
a little bit more confident
that you were right.
00:08:25.000 --> 00:08:26.438
Well, we used to argue a lot
00:08:26.438 --> 00:08:28.320
and you would sometimes
referee those.
00:08:28.320 --> 00:08:29.500
[laughter]
00:08:29.800 --> 00:08:30.800
That was fun.
00:08:32.760 --> 00:08:34.125
It wasn't hurtful.
00:08:35.200 --> 00:08:37.492
I remember it getting
a little testy once,
00:08:37.935 --> 00:08:39.606
we had lunch in The Faculty Club
00:08:40.600 --> 00:08:44.077
and we're talking about
the draft lottery paper.
00:08:44.930 --> 00:08:47.430
We were talking about "never takes"
00:08:47.430 --> 00:08:51.000
[as people who wound serve]
in the military irrespective of
00:08:51.000 --> 00:08:53.500
whether they were getting drafted
00:08:54.500 --> 00:08:58.800
and you or Don said something
about shooting yourself in the foot,
00:08:58.800 --> 00:08:59.800
[laughter]
00:08:59.800 --> 00:09:01.530
as a way of getting
out of the military
00:09:01.530 --> 00:09:03.230
and that may be
the exclusion restriction
00:09:03.230 --> 00:09:05.223
for never takes wasn't working
00:09:06.300 --> 00:09:08.520
and then the other one was going,
00:09:08.520 --> 00:09:09.791
"Well, yes you could do that
00:09:09.791 --> 00:09:12.008
but why would you want
to shoot yourself in the foot?"
00:09:12.008 --> 00:09:13.225
[laughter]
00:09:13.225 --> 00:09:15.400
It got a little [out of hand there]--
00:09:15.400 --> 00:09:17.860
I usually go for moving to Canada,
for my example,
00:09:18.690 --> 00:09:20.096
when I'm teaching that.
00:09:20.096 --> 00:09:21.365
[laughter]
00:09:22.030 --> 00:09:23.575
But things are tricky,
00:09:24.860 --> 00:09:26.595
I get students coming
from Computer Science
00:09:26.595 --> 00:09:29.943
and they want to do things
on causal inference
00:09:30.566 --> 00:09:33.460
and it takes a huge amount
of effort to figure out
00:09:33.460 --> 00:09:35.230
how they actually thinking
about problem
00:09:35.230 --> 00:09:37.000
and whether there's something there
00:09:37.000 --> 00:09:38.310
and so, now over the years,
00:09:38.310 --> 00:09:40.302
I've got a little more appreciation
for the fact
00:09:40.302 --> 00:09:41.958
that Don was actually willing to--
00:09:42.630 --> 00:09:46.000
It took him a while,
but he did engage first with Josh
00:09:46.400 --> 00:09:47.500
and then with both of us
00:09:48.380 --> 00:09:50.163
and rather than dismissing
and say,
00:09:50.163 --> 00:09:53.348
"Okay, well I can't figure out
what these guys are doing
00:09:53.348 --> 00:09:56.435
and it's probably just
not really that interesting."
00:09:57.200 --> 00:09:59.736
Everybody always wants
to figure out quickly,
00:10:00.196 --> 00:10:01.376
you want to save time
00:10:01.376 --> 00:10:03.410
and you want to save
your brain cells
00:10:03.410 --> 00:10:04.583
for other things.
00:10:05.000 --> 00:10:07.000
The fastest route to
that is to figure out
00:10:07.000 --> 00:10:08.460
why you should dismiss something.
00:10:08.460 --> 00:10:09.560
Yes.
00:10:09.560 --> 00:10:11.100
I don't need to spend time on this.
00:10:11.100 --> 00:10:12.498
♪ [music] ♪
00:10:12.498 --> 00:10:13.880
- [Narrator] If you'd like
to watch more
00:10:13.880 --> 00:10:15.822
Nobel conversations, click here,
00:10:16.220 --> 00:10:18.409
or if you'd like to learn
more about econometrics,
00:10:18.640 --> 00:10:21.240
check out Josh's "Mastering
Econometrics" series.
00:10:21.800 --> 00:10:24.540
If you'd like to learn more
about Guido, Josh, and Isaiah
00:10:24.860 --> 00:10:26.502
check out the links
in the description.
00:10:26.992 --> 00:10:28.307
♪ [music] ♪