1
00:00:00,000 --> 00:00:02,450
[SQUEAKING]

2
00:00:02,450 --> 00:00:03,430
[RUSTLING]

3
00:00:03,430 --> 00:00:05,880
[CLICKING]

4
00:00:10,072 --> 00:00:12,280
FRANK SCHILBACH: All right,
I'm going to get started.

5
00:00:12,280 --> 00:00:17,100
This is lecture 14
of 14.13 The lecture

6
00:00:17,100 --> 00:00:20,110
today is about attention.

7
00:00:20,110 --> 00:00:21,474
OK.

8
00:00:21,474 --> 00:00:22,185
Good.

9
00:00:22,185 --> 00:00:23,560
So where are we
at in this class?

10
00:00:23,560 --> 00:00:25,960
In this class, we talked a
lot about preferences so far.

11
00:00:25,960 --> 00:00:29,290
Think about preference
describing in economics.

12
00:00:29,290 --> 00:00:30,910
What do people want?

13
00:00:30,910 --> 00:00:33,015
What makes them happy
in one way or the other?

14
00:00:33,015 --> 00:00:35,140
We talked about time
preferences, risk preferences,

15
00:00:35,140 --> 00:00:37,430
and social preferences,
time preferences.

16
00:00:37,430 --> 00:00:39,410
How do people make
choices over time?

17
00:00:39,410 --> 00:00:41,380
Risk preferences
are, how do people

18
00:00:41,380 --> 00:00:44,500
deal with risky situations
when some things are uncertain

19
00:00:44,500 --> 00:00:46,600
and others are more certain?

20
00:00:46,600 --> 00:00:48,820
Or sometimes good things
happen, sometimes not.

21
00:00:48,820 --> 00:00:51,610
And there's a probability
distribution over those events.

22
00:00:51,610 --> 00:00:53,620
And how does that affect
how you make choices?

23
00:00:53,620 --> 00:00:55,120
And social
preferences is what we

24
00:00:55,120 --> 00:00:57,820
talked about during the
last few weeks, which

25
00:00:57,820 --> 00:01:01,420
is how does the presence of
others affect our choices?

26
00:01:01,420 --> 00:01:03,370
How much do we
care about others?

27
00:01:03,370 --> 00:01:06,350
And do we decide or
make choices differently

28
00:01:06,350 --> 00:01:08,270
if others are involved?

29
00:01:08,270 --> 00:01:09,910
So now what we've
done with preferences

30
00:01:09,910 --> 00:01:13,570
is sometimes saying
we have our models

31
00:01:13,570 --> 00:01:17,233
to think of ways to think
about what makes people happy.

32
00:01:17,233 --> 00:01:19,150
The second part when you
try to make decisions

33
00:01:19,150 --> 00:01:22,220
is when you look at your
environment, what are prices

34
00:01:22,220 --> 00:01:22,720
like?

35
00:01:22,720 --> 00:01:26,180
What do we know about others?

36
00:01:26,180 --> 00:01:27,160
What are our beliefs?

37
00:01:27,160 --> 00:01:28,210
And what do we attend to?

38
00:01:28,210 --> 00:01:30,170
And how do we learn when
we get new information?

39
00:01:30,170 --> 00:01:31,462
So what information do we have?

40
00:01:31,462 --> 00:01:34,180
And what information
do we attend to?

41
00:01:34,180 --> 00:01:38,420
And how do we learn when we get
new information given to us?

42
00:01:38,420 --> 00:01:41,440
And so we're going to
talk first about attention

43
00:01:41,440 --> 00:01:42,130
in this lecture.

44
00:01:42,130 --> 00:01:44,170
And then the next
couple of times, we're

45
00:01:44,170 --> 00:01:49,003
going to talk about
beliefs and learning.

46
00:01:49,003 --> 00:01:50,420
So what we're going
to do today is

47
00:01:50,420 --> 00:01:52,295
we're going to talk
about three broad issues.

48
00:01:52,295 --> 00:01:55,240
One is going to provide you
some motivating evidence

49
00:01:55,240 --> 00:01:57,580
about attention.

50
00:01:57,580 --> 00:02:00,700
We had a little bit on
this in lectures 1 and 2.

51
00:02:00,700 --> 00:02:02,805
But I wanted to remind
you of that a little bit.

52
00:02:02,805 --> 00:02:04,930
Then we're going to talk
about two specific papers.

53
00:02:04,930 --> 00:02:07,510
One is a pretty famous
paper by Raj Chetty

54
00:02:07,510 --> 00:02:10,900
and co-authors on
inattention to taxes.

55
00:02:10,900 --> 00:02:13,660
And then you're going to talk
about a second paper, which

56
00:02:13,660 --> 00:02:15,687
is learning by noticing.

57
00:02:15,687 --> 00:02:17,770
In particular, we're going
to ask the question of,

58
00:02:17,770 --> 00:02:21,430
how is it possible that people
do not attend to things that

59
00:02:21,430 --> 00:02:23,950
are important in their lives?

60
00:02:23,950 --> 00:02:26,590
You might say, well, if
attention is limited,

61
00:02:26,590 --> 00:02:27,650
that's all good.

62
00:02:27,650 --> 00:02:30,340
But then we should
expect that people attend

63
00:02:30,340 --> 00:02:31,740
to really important stuff.

64
00:02:31,740 --> 00:02:34,690
So some sort of rational
inattention models

65
00:02:34,690 --> 00:02:36,880
would say, well, it
might well be that people

66
00:02:36,880 --> 00:02:38,620
are inattentive in some ways.

67
00:02:38,620 --> 00:02:42,835
But really, it can't be
possibly that the welfare

68
00:02:42,835 --> 00:02:44,380
losses from
inattention are large

69
00:02:44,380 --> 00:02:46,300
because if they were
large, if there's things

70
00:02:46,300 --> 00:02:47,758
that people missed,
then they would

71
00:02:47,758 --> 00:02:50,530
pay attention to those things
and shift their attention

72
00:02:50,530 --> 00:02:51,260
optimally.

73
00:02:51,260 --> 00:02:54,100
The learning by
noticing paper is

74
00:02:54,100 --> 00:02:56,110
one theory, at
least potentially,

75
00:02:56,110 --> 00:03:00,070
that addresses this question.

76
00:03:00,070 --> 00:03:05,800
The way these movie perception
tests work is, essentially,

77
00:03:05,800 --> 00:03:08,210
they look for what's
called change blindness.

78
00:03:08,210 --> 00:03:10,810
So the way this works
is, often, there's

79
00:03:10,810 --> 00:03:12,910
one actor, which
is this guy here.

80
00:03:12,910 --> 00:03:15,250
I think you can see
probably right now there's

81
00:03:15,250 --> 00:03:17,800
some guy who does some activity.

82
00:03:17,800 --> 00:03:19,600
You're supposed
to watch the guy.

83
00:03:19,600 --> 00:03:24,040
And then there's a second
scene in some ways where

84
00:03:24,040 --> 00:03:26,740
the person does something else.

85
00:03:26,740 --> 00:03:29,200
In this case, the guy--

86
00:03:29,200 --> 00:03:31,030
sorry, where is he?

87
00:03:31,030 --> 00:03:32,800
The guy is taking a phone call.

88
00:03:32,800 --> 00:03:34,360
And so what happens then is--

89
00:03:37,330 --> 00:03:39,260
essentially, then, this
guy that you see here

90
00:03:39,260 --> 00:03:42,890
that takes a phone call is a
different person altogether.

91
00:03:42,890 --> 00:03:45,040
And so that's called
change blindness

92
00:03:45,040 --> 00:03:47,380
where, essentially,
people then tend

93
00:03:47,380 --> 00:03:49,480
to not notice at
all that there's two

94
00:03:49,480 --> 00:03:51,910
different people in a scene.

95
00:03:51,910 --> 00:03:53,410
There's other types
of experiments

96
00:03:53,410 --> 00:03:54,610
that do this, as well.

97
00:03:54,610 --> 00:03:57,700
One version of that would
be people who go to a bank.

98
00:03:57,700 --> 00:03:58,750
There's a bank teller.

99
00:03:58,750 --> 00:04:00,850
And this is like all
psych experiments.

100
00:04:00,850 --> 00:04:01,810
There's a bank teller.

101
00:04:04,920 --> 00:04:06,925
Here's the other guy.

102
00:04:06,925 --> 00:04:07,800
Here's the other guy.

103
00:04:07,800 --> 00:04:08,800
So you see this guy.

104
00:04:08,800 --> 00:04:10,842
He looks actually very
different to the other guy

105
00:04:10,842 --> 00:04:12,068
that you had seen before.

106
00:04:12,068 --> 00:04:13,110
So people go to the bank.

107
00:04:13,110 --> 00:04:13,830
There's a bank teller.

108
00:04:13,830 --> 00:04:15,332
The bank teller
says, oh, yeah, I'll

109
00:04:15,332 --> 00:04:17,040
have to just get
something from the back.

110
00:04:17,040 --> 00:04:19,019
And then another
person shows up who's

111
00:04:19,019 --> 00:04:21,750
a totally different person,
often looks very different,

112
00:04:21,750 --> 00:04:23,610
sometimes wears
different things.

113
00:04:23,610 --> 00:04:25,900
And people just
don't notice at all.

114
00:04:25,900 --> 00:04:27,480
And that's sort of
type of experiment

115
00:04:27,480 --> 00:04:32,030
is called change blindness.

116
00:04:32,030 --> 00:04:34,670
I've already shown you
the gorilla test before.

117
00:04:34,670 --> 00:04:37,170
I'm going to not play it again
because I've already seen it.

118
00:04:37,170 --> 00:04:38,670
But essentially,
it's the same where

119
00:04:38,670 --> 00:04:40,290
people are asked
to pay attention

120
00:04:40,290 --> 00:04:44,640
to how many passes people
play in a basketball game.

121
00:04:44,640 --> 00:04:47,100
And then there's a
guy who's in a gorilla

122
00:04:47,100 --> 00:04:53,010
suit who shows up and walks
through the whole screen.

123
00:04:53,010 --> 00:04:57,080
And many people just
miss the gorilla suit.

124
00:04:57,080 --> 00:04:59,830
Now, what's going on here?

125
00:04:59,830 --> 00:05:03,290
There is a large number of
these types of experiments.

126
00:05:03,290 --> 00:05:08,540
They're called change blindness
or inattention experiments.

127
00:05:08,540 --> 00:05:10,950
And there's another
version of that,

128
00:05:10,950 --> 00:05:13,470
which is called the dichotic
listening experiment.

129
00:05:13,470 --> 00:05:17,930
This is going back
to broadband, 1958.

130
00:05:17,930 --> 00:05:21,020
They essentially look
something like this

131
00:05:21,020 --> 00:05:25,820
where there's people who wear
headphones with messages.

132
00:05:25,820 --> 00:05:28,880
On the one hand, on
our left, there's

133
00:05:28,880 --> 00:05:30,420
supposed to be ignored inputs.

134
00:05:30,420 --> 00:05:33,020
And on the other hand, there's
supposed to be attended inputs.

135
00:05:33,020 --> 00:05:34,490
So the person in
the experiment is

136
00:05:34,490 --> 00:05:37,130
asked, only listen to
stuff that's, I guess,

137
00:05:37,130 --> 00:05:39,620
in your left ear.

138
00:05:39,620 --> 00:05:41,390
These are the attended inputs.

139
00:05:41,390 --> 00:05:44,370
I'm going to ask you some
questions about that.

140
00:05:44,370 --> 00:05:46,730
And then what you're
supposed to then do is--

141
00:05:52,447 --> 00:05:53,280
are you still stuck?

142
00:05:57,530 --> 00:05:58,220
This is bad.

143
00:06:00,720 --> 00:06:01,220
One second.

144
00:06:01,220 --> 00:06:04,790
I'm going to unshare my screen.

145
00:06:04,790 --> 00:06:05,690
Exactly.

146
00:06:05,690 --> 00:06:07,090
Give me a second.

147
00:06:07,090 --> 00:06:11,870
So the way this works is you
would have headphones on.

148
00:06:11,870 --> 00:06:15,470
And on one hand, you
hear inputs that you're

149
00:06:15,470 --> 00:06:16,740
supposed to attend to.

150
00:06:16,740 --> 00:06:18,350
On the other ear,
you hear inputs

151
00:06:18,350 --> 00:06:20,150
that you're not
supposed to attend to.

152
00:06:20,150 --> 00:06:28,810
And usually, then, you
get explicit instructions

153
00:06:28,810 --> 00:06:30,730
to attend to the
message in one ear

154
00:06:30,730 --> 00:06:33,190
and then asked later about
message in the other ear.

155
00:06:33,190 --> 00:06:35,110
People can't remember it.

156
00:06:35,110 --> 00:06:37,720
And then what these
experiments often are--

157
00:06:37,720 --> 00:06:39,520
when asked to keep a
number in their head,

158
00:06:39,520 --> 00:06:41,660
people remember the
played message much less.

159
00:06:41,660 --> 00:06:45,850
So essentially, what this
shows is that, A, I guess,

160
00:06:45,850 --> 00:06:49,240
people are pretty good at
shifting attention in some way.

161
00:06:49,240 --> 00:06:51,640
But attention is
limited to the extent

162
00:06:51,640 --> 00:06:54,190
of when you're also supposed
to attend to something

163
00:06:54,190 --> 00:06:55,750
else at the same time.

164
00:06:55,750 --> 00:06:57,580
It's essentially
too much going on,

165
00:06:57,580 --> 00:07:01,380
and it's really hard for
you to remember everything.

166
00:07:01,380 --> 00:07:05,540
And so to some degree,
it tells us, A, I guess,

167
00:07:05,540 --> 00:07:07,150
attention is limited.

168
00:07:07,150 --> 00:07:08,240
We tend to focus--

169
00:07:08,240 --> 00:07:09,680
when instructed,
at least, we tend

170
00:07:09,680 --> 00:07:11,840
to focus on stuff
that's important, right?

171
00:07:11,840 --> 00:07:13,465
So like in the
broadband experiment,

172
00:07:13,465 --> 00:07:15,590
if I tell you listen to
the stuff in the right ear,

173
00:07:15,590 --> 00:07:17,030
you actually focus
your attention

174
00:07:17,030 --> 00:07:18,350
to the stuff in the right ear.

175
00:07:18,350 --> 00:07:19,878
You miss the stuff
on the left ear,

176
00:07:19,878 --> 00:07:21,920
so you can sort of direct
your attention to stuff

177
00:07:21,920 --> 00:07:23,600
that you think is
important or you've

178
00:07:23,600 --> 00:07:25,490
been instructed to focus on.

179
00:07:25,490 --> 00:07:27,360
So that kind of works.

180
00:07:27,360 --> 00:07:30,800
But at the same time, attention
is limited in the sense of you

181
00:07:30,800 --> 00:07:33,470
can actually not remember
what's in the other ear

182
00:07:33,470 --> 00:07:34,803
because you didn't attend to it.

183
00:07:34,803 --> 00:07:36,845
So in some sense, it's
not like you automatically

184
00:07:36,845 --> 00:07:37,490
remember stuff.

185
00:07:37,490 --> 00:07:39,590
You have to pay
attention to things.

186
00:07:39,590 --> 00:07:42,680
And then at the same time,
when we now exogenously reduce

187
00:07:42,680 --> 00:07:47,150
people's attention
by telling people

188
00:07:47,150 --> 00:07:50,570
to keep a number in their head,
distracting them in some way,

189
00:07:50,570 --> 00:07:53,300
people are much more--

190
00:07:53,300 --> 00:07:55,700
even playing the played
message, the message that you're

191
00:07:55,700 --> 00:07:56,660
supposed to remember.

192
00:07:56,660 --> 00:07:59,720
I'm sure they don't remember
the other message, either.

193
00:07:59,720 --> 00:08:00,380
OK.

194
00:08:00,380 --> 00:08:04,155
So then in addition, which
I already alluded to,

195
00:08:04,155 --> 00:08:05,030
attention is limited.

196
00:08:07,678 --> 00:08:09,470
There might be different
factors that might

197
00:08:09,470 --> 00:08:10,680
affect people's attention.

198
00:08:10,680 --> 00:08:14,790
So what factors might be
important for attention?

199
00:08:14,790 --> 00:08:16,890
Could be your physical
environment in some ways.

200
00:08:16,890 --> 00:08:18,330
It could be it's really hot.

201
00:08:18,330 --> 00:08:19,730
It could be it's really cold.

202
00:08:19,730 --> 00:08:23,590
It could be other
threats in the room,

203
00:08:23,590 --> 00:08:30,870
so if you're in some
place where you're

204
00:08:30,870 --> 00:08:33,700
worried that there will be wild
animals coming anytime soon.

205
00:08:33,700 --> 00:08:36,169
Physical environment
could be really important.

206
00:08:36,169 --> 00:08:39,400
It could be really loud.

207
00:08:39,400 --> 00:08:41,750
Distractions could
be social media.

208
00:08:41,750 --> 00:08:44,560
It could be other
stuff that you're just

209
00:08:44,560 --> 00:08:46,210
trying to do at the same time.

210
00:08:46,210 --> 00:08:49,540
Maybe you're trying to solve
some problems right now.

211
00:08:49,540 --> 00:08:53,290
Maybe you're trying to prepare
for some-- for example,

212
00:08:53,290 --> 00:08:55,895
I had a phone call
just before our class.

213
00:08:55,895 --> 00:08:58,270
And I was also trying to think
about what to do in class,

214
00:08:58,270 --> 00:08:59,937
and it's just really
distracting to have

215
00:08:59,937 --> 00:09:02,350
other things you're thinking
about at the same time.

216
00:09:02,350 --> 00:09:05,260
Could be worries about your
own or others' well being.

217
00:09:05,260 --> 00:09:08,630
It could be sleep or what
people were saying already,

218
00:09:08,630 --> 00:09:11,830
people's physical environment.

219
00:09:11,830 --> 00:09:14,450
We're going to talk about some
of this in the poverty lecture.

220
00:09:14,450 --> 00:09:17,170
So there in
particular, one thing

221
00:09:17,170 --> 00:09:19,120
that poverty tends
to come along with

222
00:09:19,120 --> 00:09:21,800
is sort of a bunch of
different ways of--

223
00:09:21,800 --> 00:09:25,840
in addition to not having money
and many other deprivations,

224
00:09:25,840 --> 00:09:30,790
poverty tends to come along
with lots of other factors

225
00:09:30,790 --> 00:09:33,190
that might reduce
people's attention.

226
00:09:33,190 --> 00:09:34,780
And that's to do with sleep.

227
00:09:34,780 --> 00:09:37,360
It's to do with
exposure to noise.

228
00:09:37,360 --> 00:09:40,990
It's coming with worries
about people's health

229
00:09:40,990 --> 00:09:42,440
and so on and so forth.

230
00:09:42,440 --> 00:09:44,350
There's some question
about practice.

231
00:09:44,350 --> 00:09:45,940
Can you actually
practice attention?

232
00:09:45,940 --> 00:09:47,315
And there's some
people who would

233
00:09:47,315 --> 00:09:49,870
argue, at least in education,
maybe at younger ages--

234
00:09:49,870 --> 00:09:52,240
suppose you sit
kids down and have

235
00:09:52,240 --> 00:09:55,660
them practice focusing on
tasks like multiplications

236
00:09:55,660 --> 00:09:57,590
or whatever and
do math problems.

237
00:09:57,590 --> 00:09:59,950
Some of this, some
people would argue,

238
00:09:59,950 --> 00:10:04,840
in addition to
just helping people

239
00:10:04,840 --> 00:10:08,980
learn math, actually the numbers
and so on and how to multiply,

240
00:10:08,980 --> 00:10:11,380
you might also
actually just help

241
00:10:11,380 --> 00:10:15,130
students learn how to focus and
how to pay attention carefully

242
00:10:15,130 --> 00:10:18,640
to things and particularly
sustained attention

243
00:10:18,640 --> 00:10:22,660
and not get distracted
by other things.

244
00:10:22,660 --> 00:10:24,250
Other distractions,
by the way, also,

245
00:10:24,250 --> 00:10:25,730
like temptations, for example.

246
00:10:25,730 --> 00:10:29,290
So if I put a delicious
chocolate cake in front of you

247
00:10:29,290 --> 00:10:31,323
and have you-- and if
you're really hungry,

248
00:10:31,323 --> 00:10:32,740
that might be
actually really hard

249
00:10:32,740 --> 00:10:34,180
to pay attention
to whatever else

250
00:10:34,180 --> 00:10:38,380
you're doing just because that's
really distracting to you.

251
00:10:38,380 --> 00:10:39,080
OK.

252
00:10:39,080 --> 00:10:43,750
And so we're going to,
for now, keep thinking

253
00:10:43,750 --> 00:10:45,038
about attention being fixed.

254
00:10:45,038 --> 00:10:47,080
We're going to get back
to the issue of attention

255
00:10:47,080 --> 00:10:50,080
being malleable and
potentially being

256
00:10:50,080 --> 00:10:54,250
depleted by issues such as
poverty at the end in the very

257
00:10:54,250 --> 00:10:54,910
last lecture.

258
00:10:54,910 --> 00:10:56,500
But for now, we're
going to look at--

259
00:10:56,500 --> 00:10:59,230
think about attention
as fixed but limited.

260
00:10:59,230 --> 00:11:00,730
And now we're going
to look at, what

261
00:11:00,730 --> 00:11:04,670
are the consequences of that?

262
00:11:04,670 --> 00:11:05,263
OK.

263
00:11:05,263 --> 00:11:06,680
So then one question
you might ask

264
00:11:06,680 --> 00:11:09,340
is, well, this listening
task that I just showed you

265
00:11:09,340 --> 00:11:12,440
or when a monkey walks through
some video or the like,

266
00:11:12,440 --> 00:11:14,240
it's a pretty
low-stakes situation.

267
00:11:14,240 --> 00:11:15,600
Who really cares?

268
00:11:15,600 --> 00:11:17,630
And one Chicago
view of economics

269
00:11:17,630 --> 00:11:21,560
would be to say, well, these
are just low-stakes situations.

270
00:11:21,560 --> 00:11:25,670
But once stakes are
really high, then people

271
00:11:25,670 --> 00:11:27,298
will surely pay attention.

272
00:11:27,298 --> 00:11:29,090
And so there's a number
of different papers

273
00:11:29,090 --> 00:11:30,500
that look at this issue.

274
00:11:30,500 --> 00:11:39,380
But one quite nice example
here is from the stock market,

275
00:11:39,380 --> 00:11:42,290
arguably an event of
high economic importance

276
00:11:42,290 --> 00:11:45,230
because people make a lot of
money, potentially, from it.

277
00:11:45,230 --> 00:11:47,480
So there's this very nice
people by Huberman and Regev

278
00:11:47,480 --> 00:11:51,110
from 2001 which has--

279
00:11:51,110 --> 00:11:53,660
it's essentially an event
study and a case study

280
00:11:53,660 --> 00:11:59,810
of stock market development
of a company called EntreMed.

281
00:11:59,810 --> 00:12:04,730
And this company in 1997
on October, November

282
00:12:04,730 --> 00:12:08,360
has very positive early
results on a cure for cancer.

283
00:12:08,360 --> 00:12:10,760
Now, for any of these
types of companies,

284
00:12:10,760 --> 00:12:13,550
having new cures and new
medications or vaccines

285
00:12:13,550 --> 00:12:16,760
or the like are hugely valuable
because they can sell a lot

286
00:12:16,760 --> 00:12:20,660
of their product in the future,
which means, essentially,

287
00:12:20,660 --> 00:12:24,170
positive news about that
means you'll have lots--

288
00:12:24,170 --> 00:12:28,790
at least with high probability
or higher probability--

289
00:12:28,790 --> 00:12:31,310
large profits in the
future, to the extent

290
00:12:31,310 --> 00:12:35,430
that the stock market is
an indication of that.

291
00:12:35,430 --> 00:12:39,020
You might expect the
stock market to go up.

292
00:12:39,020 --> 00:12:42,470
Now then, November 28,
the journal Nature,

293
00:12:42,470 --> 00:12:47,120
which was a very highly
prominent scientific journal,

294
00:12:47,120 --> 00:12:50,150
features this new
very prominently.

295
00:12:50,150 --> 00:12:52,560
And The New York Times,
in fact, reports on it

296
00:12:52,560 --> 00:12:57,350
on page A28, which is very
much in the back of the paper.

297
00:12:57,350 --> 00:13:02,240
And on May 3, this is a
whole six months later.

298
00:13:02,240 --> 00:13:04,130
The New York Times
again features

299
00:13:04,130 --> 00:13:09,130
essentially the same story as on
November 28 on the front page.

300
00:13:09,130 --> 00:13:10,250
OK.

301
00:13:10,250 --> 00:13:17,600
Now, then on November 12, 1998,
the Wall Street Journal front

302
00:13:17,600 --> 00:13:22,040
page then reports about
a failed replication.

303
00:13:22,040 --> 00:13:23,600
It's essentially
to say, well, these

304
00:13:23,600 --> 00:13:26,060
are promising new early results.

305
00:13:26,060 --> 00:13:28,357
These early results are
often small samples.

306
00:13:28,357 --> 00:13:30,440
And it's not quite clear,
does it replicate really

307
00:13:30,440 --> 00:13:31,940
with larger samples and so on.

308
00:13:31,940 --> 00:13:33,740
Is this really a good cure?

309
00:13:33,740 --> 00:13:36,710
And so you kind of
won't have replications.

310
00:13:36,710 --> 00:13:40,160
And really, only if you're
able to replicate your results

311
00:13:40,160 --> 00:13:41,870
is this actually
something that's useful

312
00:13:41,870 --> 00:13:44,590
and that you can make money off.

313
00:13:44,590 --> 00:13:48,260
Now, what happened to the
EntreMed stock prices?

314
00:13:48,260 --> 00:13:50,120
And so the first
question you might

315
00:13:50,120 --> 00:13:52,310
say is, well, in a
world of full attention

316
00:13:52,310 --> 00:13:54,860
with unlimited arbitrage,
as in if people

317
00:13:54,860 --> 00:13:58,400
took full advantage of the
news and reacted perfectly,

318
00:13:58,400 --> 00:13:59,660
what would we expect?

319
00:13:59,660 --> 00:14:03,190
And second, in reality,
what did actually happen?

320
00:14:03,190 --> 00:14:06,590
So let's first ask the
first question about,

321
00:14:06,590 --> 00:14:10,013
with full attention,
what would we expect?

322
00:14:10,013 --> 00:14:11,680
I guess there's a bit
of a question what

323
00:14:11,680 --> 00:14:12,770
happens in the first one.

324
00:14:12,770 --> 00:14:15,020
There's a bit of a question,
is there insider trading?

325
00:14:15,020 --> 00:14:16,747
Or who gets the news
in the first place?

326
00:14:16,747 --> 00:14:19,330
So one thing you might say is,
well, maybe the first one, only

327
00:14:19,330 --> 00:14:20,350
the company knows.

328
00:14:20,350 --> 00:14:23,170
And maybe that's not
public knowledge.

329
00:14:23,170 --> 00:14:24,700
And then if there's
insider trading,

330
00:14:24,700 --> 00:14:26,650
maybe the stock
market might go up.

331
00:14:26,650 --> 00:14:29,463
But surely, like in number
two, if Nature prominently

332
00:14:29,463 --> 00:14:31,630
features it, there's going
to be a bunch of traders,

333
00:14:31,630 --> 00:14:34,242
a bunch of people who invest
in this company or not.

334
00:14:34,242 --> 00:14:35,950
If they pay attention,
they should really

335
00:14:35,950 --> 00:14:38,260
know what's going on,
and they should really--

336
00:14:38,260 --> 00:14:40,600
the stock market should go up.

337
00:14:40,600 --> 00:14:44,980
Conditional on that, on the
November 28 news, on May 3,

338
00:14:44,980 --> 00:14:46,817
there's really no
news here, right?

339
00:14:46,817 --> 00:14:48,400
It's kind of like
we knew this before.

340
00:14:48,400 --> 00:14:51,210
If everybody was paying
attention before carefully,

341
00:14:51,210 --> 00:14:52,960
the stock market should
have incorporated.

342
00:14:52,960 --> 00:14:54,250
The news is out.

343
00:14:54,250 --> 00:14:56,020
Traders know.

344
00:14:56,020 --> 00:15:00,280
You should buy this, or
stocks at this company

345
00:15:00,280 --> 00:15:01,662
is going to be more profitable.

346
00:15:01,662 --> 00:15:02,620
So stocks should go up.

347
00:15:02,620 --> 00:15:06,250
So really, if there's
full attention,

348
00:15:06,250 --> 00:15:10,030
you expect nothing to
happen on May 3, '98.

349
00:15:10,030 --> 00:15:12,340
And then if there's a
failed replication, well,

350
00:15:12,340 --> 00:15:18,190
to the extent that this means
this cancer cure is entirely

351
00:15:18,190 --> 00:15:21,520
useless, the stock market
should essentially just go down

352
00:15:21,520 --> 00:15:24,530
to where it was before.

353
00:15:24,530 --> 00:15:28,400
Now, but what about in reality?

354
00:15:28,400 --> 00:15:31,890
So let me actually
just show you.

355
00:15:31,890 --> 00:15:33,940
So here's the reality of this.

356
00:15:33,940 --> 00:15:36,170
So there's not even--

357
00:15:36,170 --> 00:15:39,410
there's the previous
events done in here.

358
00:15:39,410 --> 00:15:42,300
This is essentially the
first story that comes out.

359
00:15:42,300 --> 00:15:44,150
You see a little bit
of an increase here.

360
00:15:44,150 --> 00:15:46,567
But actually, quite a bit in
relative terms with something

361
00:15:46,567 --> 00:15:48,050
like 5%.

362
00:15:48,050 --> 00:15:50,210
It seems like people
just didn't really quite

363
00:15:50,210 --> 00:15:52,370
notice what was going on.

364
00:15:52,370 --> 00:15:55,137
And there's also
no trend upwards.

365
00:15:55,137 --> 00:15:57,470
It's not like the news is
spreading across people and so

366
00:15:57,470 --> 00:15:58,160
on.

367
00:15:58,160 --> 00:16:01,470
And then there's like a huge
jump now in The New York Times,

368
00:16:01,470 --> 00:16:01,970
right?

369
00:16:01,970 --> 00:16:04,370
So this is essentially here,
it's in The New York Times

370
00:16:04,370 --> 00:16:06,410
but in the back of the journal.

371
00:16:06,410 --> 00:16:09,080
Up here, it's essentially The
New York Times, the front page.

372
00:16:09,080 --> 00:16:11,930
And suddenly, lots of
people pay attention

373
00:16:11,930 --> 00:16:15,760
to news that's actually not
news because that was already

374
00:16:15,760 --> 00:16:17,000
in The New York Times before.

375
00:16:17,000 --> 00:16:18,792
To the extent that
everybody pays attention

376
00:16:18,792 --> 00:16:21,140
to this, unlimited attention
and unlimited arbitrage,

377
00:16:21,140 --> 00:16:24,420
we should essentially expect
no effects whatsoever.

378
00:16:24,420 --> 00:16:27,170
So we see a huge spike in
the stock market price.

379
00:16:27,170 --> 00:16:29,420
Maybe the stock market
overreacted a bit

380
00:16:29,420 --> 00:16:31,080
and went down a little bit.

381
00:16:31,080 --> 00:16:33,770
But then what we see here is
then the negative effective

382
00:16:33,770 --> 00:16:35,510
of the failed replication.

383
00:16:35,510 --> 00:16:37,910
Notice that here, the stock
market is still quite a bit

384
00:16:37,910 --> 00:16:38,150
high.

385
00:16:38,150 --> 00:16:40,650
The price is still up quite a
bit high, maybe twice as high,

386
00:16:40,650 --> 00:16:44,115
about 20 compared to
10 as it was before.

387
00:16:44,115 --> 00:16:46,490
It's a little bit hard to
interpret that because it could

388
00:16:46,490 --> 00:16:50,330
just be that maybe
there's still a chance

389
00:16:50,330 --> 00:16:52,040
that the cure might work out.

390
00:16:52,040 --> 00:16:54,500
Maybe there's other news about
the company in the meantime

391
00:16:54,500 --> 00:16:56,130
that are positive and so on.

392
00:16:56,130 --> 00:16:58,430
But the key part
here is the fact

393
00:16:58,430 --> 00:17:02,000
that the stock market went
up here on May 4, 1998,

394
00:17:02,000 --> 00:17:07,974
really is saying it's
news for lots of people.

395
00:17:07,974 --> 00:17:09,349
And really, it
shouldn't be news.

396
00:17:09,349 --> 00:17:10,849
And there shouldn't
be any reaction

397
00:17:10,849 --> 00:17:12,650
because, essentially,
it's old news,

398
00:17:12,650 --> 00:17:16,700
as you would say, in the
sense of people in 1997

399
00:17:16,700 --> 00:17:19,745
in November 28 already
reported all of that.

400
00:17:22,339 --> 00:17:25,569
Does that make sense?

401
00:17:25,569 --> 00:17:27,140
OK.

402
00:17:27,140 --> 00:17:29,390
And there's lots of examples
of these kinds of things.

403
00:17:29,390 --> 00:17:32,020
But essentially,
some of this is--

404
00:17:32,020 --> 00:17:34,288
in attention, some of
it is also confusion.

405
00:17:34,288 --> 00:17:35,830
I don't know if any
of you have heard

406
00:17:35,830 --> 00:17:40,170
of what happened to the
Zoom stock market price,

407
00:17:40,170 --> 00:17:40,920
stock price.

408
00:17:44,083 --> 00:17:46,250
I don't know if any of you
have heard of this story.

409
00:17:46,250 --> 00:17:49,000
So essentially,
there's a company Zoom,

410
00:17:49,000 --> 00:17:50,650
which is the company
that we're using

411
00:17:50,650 --> 00:17:53,440
right now to record this video.

412
00:17:53,440 --> 00:17:55,630
The stock market
price of Zoom has

413
00:17:55,630 --> 00:17:57,820
gone up a lot during
the last three months,

414
00:17:57,820 --> 00:18:00,010
rightly so because
Zoom now has--

415
00:18:00,010 --> 00:18:02,190
Zoom's profits
has gone up a lot.

416
00:18:02,190 --> 00:18:04,345
For example, MIT now
has a corporate license

417
00:18:04,345 --> 00:18:07,630
of Zoom, which essentially
means thousands of people

418
00:18:07,630 --> 00:18:10,300
now are using or can use Zoom,
and that's an expensive thing

419
00:18:10,300 --> 00:18:12,460
to do.

420
00:18:12,460 --> 00:18:15,640
It turns out there's
another company that's

421
00:18:15,640 --> 00:18:18,790
also called Zoom except for
that it's a different company

422
00:18:18,790 --> 00:18:21,460
and has nothing to do with
the actual Zoom technology.

423
00:18:21,460 --> 00:18:23,650
It turns out that
the Zoom stock market

424
00:18:23,650 --> 00:18:26,650
price for this other company
went up a lot, as well.

425
00:18:26,650 --> 00:18:31,870
The ticker for that company is
actually ZOOM, so like Zoom,

426
00:18:31,870 --> 00:18:34,030
as opposed to the
actual Zoom company

427
00:18:34,030 --> 00:18:35,200
that we're using right now.

428
00:18:35,200 --> 00:18:37,210
Their ticker is ZM.

429
00:18:37,210 --> 00:18:39,940
And so at some point,
actually, the trading

430
00:18:39,940 --> 00:18:42,393
for that other company
was halted, presumably

431
00:18:42,393 --> 00:18:43,810
because people
were very confused,

432
00:18:43,810 --> 00:18:45,768
and they were not really
paying close attention

433
00:18:45,768 --> 00:18:48,512
to what is the right ticker,
what's the right company

434
00:18:48,512 --> 00:18:49,720
that they're actually buying.

435
00:18:49,720 --> 00:18:51,178
They thought they
were buying Zoom,

436
00:18:51,178 --> 00:18:55,090
like the actual online
video conferencing

437
00:18:55,090 --> 00:18:56,695
company, while in
fact they were just

438
00:18:56,695 --> 00:18:58,570
buying some other company
that's actually not

439
00:18:58,570 --> 00:19:00,610
even doing anything anymore.

440
00:19:00,610 --> 00:19:05,800
So there's lots of stories
about people being inattentive

441
00:19:05,800 --> 00:19:10,530
even in situations when stakes
are really, really high.

442
00:19:10,530 --> 00:19:13,350
Now, then, you might say,
well, how do we now measure?

443
00:19:13,350 --> 00:19:16,980
So now in some sense we
have some evidence here

444
00:19:16,980 --> 00:19:19,620
that people are-- in some sense
somewhat informal evidence

445
00:19:19,620 --> 00:19:21,660
that people are inattentive.

446
00:19:21,660 --> 00:19:25,950
Now, how would you measure the
impact of inattention overall?

447
00:19:28,490 --> 00:19:30,730
So if you wanted to just
measure and demonstrate

448
00:19:30,730 --> 00:19:33,340
that people are inattentive,
in addition to what I've just

449
00:19:33,340 --> 00:19:34,690
shown you, what might you do?

450
00:19:34,690 --> 00:19:36,960
How might you do that?

451
00:19:36,960 --> 00:19:40,100
There's some information that
you usually get in the world

452
00:19:40,100 --> 00:19:41,600
and they should have.

453
00:19:41,600 --> 00:19:45,560
One example that we can discuss
is essentially prices or taxes.

454
00:19:45,560 --> 00:19:48,320
And now if you want to
demonstrate that people

455
00:19:48,320 --> 00:19:50,720
are actually missing those
prices or taxes, what

456
00:19:50,720 --> 00:19:53,210
you can do is now you
can make it very salient.

457
00:19:53,210 --> 00:19:55,220
You make it really
apparent to people.

458
00:19:55,220 --> 00:20:00,320
And then you can look at, when
you look at changes in prices

459
00:20:00,320 --> 00:20:03,680
or the change of making things
from non-selling to selling,

460
00:20:03,680 --> 00:20:04,940
how do people react?

461
00:20:04,940 --> 00:20:07,640
Or maybe you make, for
example, taxes very salient.

462
00:20:07,640 --> 00:20:11,240
What happens to people's
demand for certain products?

463
00:20:11,240 --> 00:20:15,170
And if then making things
salient changes people's

464
00:20:15,170 --> 00:20:20,030
behavior, that then identifies
underlying inattention.

465
00:20:20,030 --> 00:20:21,860
So if you just-- people
don't have access

466
00:20:21,860 --> 00:20:23,480
to certain information,
in some sense, that's

467
00:20:23,480 --> 00:20:25,188
less interesting
because that just means,

468
00:20:25,188 --> 00:20:27,440
well, I don't have
the information

469
00:20:27,440 --> 00:20:29,200
that's really available to me.

470
00:20:29,200 --> 00:20:30,950
But in some ways, I
could make information

471
00:20:30,950 --> 00:20:32,330
that's really very salient--

472
00:20:32,330 --> 00:20:34,700
I essentially could
reduce the salience,

473
00:20:34,700 --> 00:20:37,250
as you say, of certain types
of information-- again,

474
00:20:37,250 --> 00:20:39,545
information that should
be available to them.

475
00:20:39,545 --> 00:20:43,340
For example, suppose I do sales.

476
00:20:43,340 --> 00:20:46,010
And I'm trying to, as
a company, sell things.

477
00:20:46,010 --> 00:20:46,850
You go to a store.

478
00:20:46,850 --> 00:20:49,770
And usually, sales are made
very salient to people.

479
00:20:49,770 --> 00:20:51,770
Now you could essentially
just change the prices

480
00:20:51,770 --> 00:20:53,930
without making these
sales very salient.

481
00:20:53,930 --> 00:20:57,680
And then you could look at, how
attentive are people to that?

482
00:20:57,680 --> 00:21:00,470
In a way, that's, in some
ways, less natural, right,

483
00:21:00,470 --> 00:21:03,248
because when people are--

484
00:21:03,248 --> 00:21:04,790
in a way, what we
want to demonstrate

485
00:21:04,790 --> 00:21:08,390
is that in people's
real lives, there's

486
00:21:08,390 --> 00:21:10,730
information available to them.

487
00:21:10,730 --> 00:21:13,940
And they're not really paying
sufficiently much attention

488
00:21:13,940 --> 00:21:15,000
to it.

489
00:21:15,000 --> 00:21:18,200
So if you can then
change that in some ways

490
00:21:18,200 --> 00:21:20,670
and demonstrate that now
they're changing their behavior,

491
00:21:20,670 --> 00:21:22,220
that must mean that
previously, they

492
00:21:22,220 --> 00:21:25,700
had been like misoptimizing.

493
00:21:25,700 --> 00:21:28,280
If instead you
did something like

494
00:21:28,280 --> 00:21:34,340
if you reduce the salience of
certain types of information,

495
00:21:34,340 --> 00:21:36,840
that's, in a way, somewhat less
interesting because it would

496
00:21:36,840 --> 00:21:39,898
say, well, it doesn't
really mean that they

497
00:21:39,898 --> 00:21:41,190
were inattentive to start with.

498
00:21:41,190 --> 00:21:42,930
It's more like you're hiding
something from people,

499
00:21:42,930 --> 00:21:43,990
and now they can't find it.

500
00:21:43,990 --> 00:21:45,990
That's, in some ways,
somewhat less interesting.

501
00:21:45,990 --> 00:21:49,230
But I think in principle,
you'd also, exactly as you say,

502
00:21:49,230 --> 00:21:51,570
identify inattention.

503
00:21:51,570 --> 00:21:55,060
So there's a few type of
things that people have done.

504
00:21:55,060 --> 00:21:57,660
And to be clear, there's
often a pretty fine line

505
00:21:57,660 --> 00:21:59,778
between attention and memory.

506
00:21:59,778 --> 00:22:02,070
So one thing we could do is--
and this is what Maya was

507
00:22:02,070 --> 00:22:03,840
saying previously--

508
00:22:03,840 --> 00:22:07,090
you could make certain features,
like taxes, very salient.

509
00:22:07,090 --> 00:22:10,010
That's what we're
going to look at next.

510
00:22:10,010 --> 00:22:13,290
In addition, you could
provide some information

511
00:22:13,290 --> 00:22:15,420
when the correct response
is known already.

512
00:22:15,420 --> 00:22:17,378
That's one we're going
to talk about afterwards

513
00:22:17,378 --> 00:22:19,020
is the study by Hanna et al.

514
00:22:19,020 --> 00:22:21,150
So that's kind of
like when people

515
00:22:21,150 --> 00:22:23,310
have certain types of
information already

516
00:22:23,310 --> 00:22:26,130
but then providing that
information again or giving

517
00:22:26,130 --> 00:22:28,920
that information in some
sort of concise form

518
00:22:28,920 --> 00:22:32,670
and potentially notifying
people of the fact

519
00:22:32,670 --> 00:22:36,190
that this information could
be or should be important.

520
00:22:36,190 --> 00:22:38,310
And then we now see
people react to it.

521
00:22:38,310 --> 00:22:41,100
That doesn't mean that that
must mean that, before, they

522
00:22:41,100 --> 00:22:42,480
were inattentive.

523
00:22:42,480 --> 00:22:45,150
Another thing that people
have done a lot is reminders.

524
00:22:45,150 --> 00:22:47,972
Reminders often
are sort of like--

525
00:22:47,972 --> 00:22:49,930
you can think of the
pieces like memory issues,

526
00:22:49,930 --> 00:22:51,630
and people just forget things.

527
00:22:51,630 --> 00:22:54,120
But in some sense, memory
issues and attention issues

528
00:22:54,120 --> 00:22:55,505
are very closely linked.

529
00:22:55,505 --> 00:22:56,880
If you forget
something, then you

530
00:22:56,880 --> 00:22:58,570
don't pay attention
to something.

531
00:22:58,570 --> 00:23:00,450
So if I remind you of
something, now you're

532
00:23:00,450 --> 00:23:02,040
paying attention
to, for example,

533
00:23:02,040 --> 00:23:05,340
your savings or your medical
adherence, like you should

534
00:23:05,340 --> 00:23:07,380
take certain drugs and so on.

535
00:23:07,380 --> 00:23:10,050
There's a number of studies that
essentially provide reminders

536
00:23:10,050 --> 00:23:11,370
to people.

537
00:23:11,370 --> 00:23:13,800
And if these reminders
have effects,

538
00:23:13,800 --> 00:23:16,620
then that must mean that people
were not paying attention

539
00:23:16,620 --> 00:23:19,470
to those kinds of
things, often presumably

540
00:23:19,470 --> 00:23:20,760
because of memory problems.

541
00:23:20,760 --> 00:23:24,330
They just forget sometimes.

542
00:23:24,330 --> 00:23:25,140
OK.

543
00:23:25,140 --> 00:23:29,010
So now let me start
with a very simple model

544
00:23:29,010 --> 00:23:33,030
to help us understand the
results from the Chetty et al.

545
00:23:33,030 --> 00:23:33,690
paper.

546
00:23:33,690 --> 00:23:37,813
It's very simple and very
sort of just sketched.

547
00:23:37,813 --> 00:23:39,480
But I think it's
actually helpful to see

548
00:23:39,480 --> 00:23:40,900
what people are doing.

549
00:23:40,900 --> 00:23:43,350
So consider a good
with a value of V

550
00:23:43,350 --> 00:23:47,190
inclusive of the price, which
is the sum of two components.

551
00:23:47,190 --> 00:23:50,430
The components are
little v and o.

552
00:23:50,430 --> 00:23:53,250
So there's visible and
salient component v,

553
00:23:53,250 --> 00:23:56,040
and there's an
opaque component o.

554
00:23:56,040 --> 00:23:56,820
OK.

555
00:23:56,820 --> 00:24:00,750
And so if you're
now inattentive,

556
00:24:00,750 --> 00:24:05,130
the consumer, inattentive
consumer perceived

557
00:24:05,130 --> 00:24:08,040
value-- instead of
perceiving the true value V,

558
00:24:08,040 --> 00:24:11,010
the consumer perceives
the value of V hat,

559
00:24:11,010 --> 00:24:13,815
which is little v, which
is a salient component.

560
00:24:13,815 --> 00:24:15,690
So even if you're
inattentive, the assumption

561
00:24:15,690 --> 00:24:20,290
is the salient part of
the good you always see.

562
00:24:20,290 --> 00:24:22,290
For example, what color
is the good or the like?

563
00:24:22,290 --> 00:24:24,466
You always see that.

564
00:24:24,466 --> 00:24:28,440
And in addition, the opaque
component, the stuff that's

565
00:24:28,440 --> 00:24:31,050
not salient, you're only
going to pay attention

566
00:24:31,050 --> 00:24:35,810
to the fraction
of 1 minus theta.

567
00:24:35,810 --> 00:24:37,890
So the degree of
inattention-- so theta

568
00:24:37,890 --> 00:24:40,340
is our inattention parameter.

569
00:24:40,340 --> 00:24:43,230
And theta measures how much
are you paying attention

570
00:24:43,230 --> 00:24:46,020
to stuff that's not
particularly salient.

571
00:24:46,020 --> 00:24:48,950
So for theta equals 0, you're
back to the cases before.

572
00:24:48,950 --> 00:24:53,640
For theta equals 0, V hat
equals V, which is essentially

573
00:24:53,640 --> 00:24:55,360
just a standard case.

574
00:24:55,360 --> 00:24:58,740
If theta were 1-- so theta is
supposed to be between 0 and 1.

575
00:24:58,740 --> 00:25:00,570
If theta is 1, then
essentially you're

576
00:25:00,570 --> 00:25:04,490
not paying attention at all
to the opaque component.

577
00:25:04,490 --> 00:25:06,780
And anywhere in between,
essentially, then, you're

578
00:25:06,780 --> 00:25:10,630
paying only partial attention
to the opaque component.

579
00:25:10,630 --> 00:25:16,150
So the interpretation is each
individual essentially sees v--

580
00:25:16,150 --> 00:25:19,770
sorry, sees o to some degree
but processes it only partially

581
00:25:19,770 --> 00:25:22,620
to the degree of theta.

582
00:25:22,620 --> 00:25:25,740
Of course, whether you actually
see it and not process it

583
00:25:25,740 --> 00:25:28,260
or whether you miss it
entirely is, in some sense,

584
00:25:28,260 --> 00:25:29,470
a philosophical question.

585
00:25:29,470 --> 00:25:32,180
But think about like, everybody
has, at least in principle,

586
00:25:32,180 --> 00:25:34,080
access to the component o.

587
00:25:34,080 --> 00:25:38,700
But they only process
and pay attention to it

588
00:25:38,700 --> 00:25:41,610
only to some degree.

589
00:25:41,610 --> 00:25:42,270
OK.

590
00:25:42,270 --> 00:25:47,410
So Chetty et al. applies this
model essentially to taxes.

591
00:25:47,410 --> 00:25:51,150
So one very interesting
feature about some taxes

592
00:25:51,150 --> 00:25:55,080
is the fact that sales taxes
are only added at the register,

593
00:25:55,080 --> 00:25:56,670
right?

594
00:25:56,670 --> 00:26:03,300
At least in most cases, you
would shop in the store.

595
00:26:03,300 --> 00:26:05,910
And then the sales tax, only
at the end once you actually

596
00:26:05,910 --> 00:26:10,620
go to the store, will be added.

597
00:26:10,620 --> 00:26:13,890
Now, people are not--

598
00:26:13,890 --> 00:26:14,400
people know.

599
00:26:14,400 --> 00:26:15,838
When you ask people
directly, what

600
00:26:15,838 --> 00:26:18,255
is the sales tax in your state,
people actually on average

601
00:26:18,255 --> 00:26:19,588
are pretty good at knowing this.

602
00:26:19,588 --> 00:26:22,510
So it's not like people don't
know that there's a sales tax.

603
00:26:22,510 --> 00:26:24,900
But they might just
forget that there

604
00:26:24,900 --> 00:26:28,380
are sales taxes for goods.

605
00:26:28,380 --> 00:26:32,550
So now what you can do now
is you can compare the demand

606
00:26:32,550 --> 00:26:38,340
response to sales tax changes
versus the demand response

607
00:26:38,340 --> 00:26:40,050
to other price changes, right?

608
00:26:40,050 --> 00:26:44,640
So if you see prices
fluctuating in a store over time

609
00:26:44,640 --> 00:26:48,240
or across goods,
presumably, everybody

610
00:26:48,240 --> 00:26:50,280
sees those price
changes overall,

611
00:26:50,280 --> 00:26:53,160
if these are price
changes that are not

612
00:26:53,160 --> 00:26:55,050
related to sales
taxes, the reason

613
00:26:55,050 --> 00:26:58,257
being once you're in the store,
you're going to look at prices.

614
00:26:58,257 --> 00:27:00,090
You're going to see is
the price high or low

615
00:27:00,090 --> 00:27:01,560
even before you go
to the register.

616
00:27:01,560 --> 00:27:05,160
So presumably, you see the
price tag pretty much most

617
00:27:05,160 --> 00:27:06,310
of the time.

618
00:27:06,310 --> 00:27:09,900
In contrast, when you
look at sales tax changes,

619
00:27:09,900 --> 00:27:12,750
if the sales taxes
go up for some reason

620
00:27:12,750 --> 00:27:16,290
or if there's a sales tax
for some items versus others,

621
00:27:16,290 --> 00:27:19,410
you might just miss
that at the time

622
00:27:19,410 --> 00:27:22,350
when you're choosing your goods
before you go to the register.

623
00:27:22,350 --> 00:27:24,120
And once you're then
at the register,

624
00:27:24,120 --> 00:27:27,810
you might not even notice that
you have paid the sales tax.

625
00:27:27,810 --> 00:27:29,490
Or you might be
surprised and just don't

626
00:27:29,490 --> 00:27:34,540
change your behavior any
more but at least notice it.

627
00:27:34,540 --> 00:27:36,010
So now what did
Chetty et al. do?

628
00:27:36,010 --> 00:27:39,610
They have data on the demand
for items in a grocery store.

629
00:27:39,610 --> 00:27:44,830
And they have essentially
the demand, D of V hat.

630
00:27:44,830 --> 00:27:47,648
Remember, V hat is the function
of the perceived value,

631
00:27:47,648 --> 00:27:48,940
which I showed you here before.

632
00:27:48,940 --> 00:27:50,340
Let me go back for a second.

633
00:27:50,340 --> 00:27:52,450
So you see V hat
here is v, which

634
00:27:52,450 --> 00:27:55,750
is the salient component,
plus 1 minus theta times o

635
00:27:55,750 --> 00:27:57,710
as the opaque component.

636
00:27:57,710 --> 00:28:02,380
So the demand for goods depends
on the perceived value, V hat,

637
00:28:02,380 --> 00:28:03,890
that people have.

638
00:28:03,890 --> 00:28:07,600
And there's a visible
part as the value V, which

639
00:28:07,600 --> 00:28:09,500
is inclusive of the price.

640
00:28:09,500 --> 00:28:11,735
So when you look at
when you're in a store,

641
00:28:11,735 --> 00:28:12,610
you look at the good.

642
00:28:12,610 --> 00:28:13,300
You see the good.

643
00:28:13,300 --> 00:28:14,467
You see whether you like it.

644
00:28:14,467 --> 00:28:15,340
You see the brand.

645
00:28:15,340 --> 00:28:17,600
You see the color and
so on and so forth.

646
00:28:17,600 --> 00:28:20,680
You also see the
price, so the valuation

647
00:28:20,680 --> 00:28:23,592
that you give or the value
V that you draw from that

648
00:28:23,592 --> 00:28:25,300
is essentially telling
how much they want

649
00:28:25,300 --> 00:28:29,860
to have this thing minus
the price that you see,

650
00:28:29,860 --> 00:28:32,890
which goes into the visible
part of your valuation.

651
00:28:32,890 --> 00:28:34,670
And then there's a
less visible part,

652
00:28:34,670 --> 00:28:36,820
which is just stuff
that's essentially hidden.

653
00:28:36,820 --> 00:28:39,430
And there's a bunch of
different parts to that

654
00:28:39,430 --> 00:28:41,320
that you might not
pay attention to,

655
00:28:41,320 --> 00:28:44,260
which could be
ingredients or the like,

656
00:28:44,260 --> 00:28:46,460
or was it organic
or not and the like.

657
00:28:46,460 --> 00:28:48,670
But in particular,
the sales tax, right?

658
00:28:48,670 --> 00:28:51,190
So if you think
about your valuation,

659
00:28:51,190 --> 00:28:52,810
you would have to
essentially, however

660
00:28:52,810 --> 00:28:54,880
much you like the
good-- you have

661
00:28:54,880 --> 00:28:57,850
to subtract the
sales tax because you

662
00:28:57,850 --> 00:29:00,290
have to pay taxes on that good.

663
00:29:00,290 --> 00:29:02,990
And that's less
visible to some people.

664
00:29:02,990 --> 00:29:05,660
So you might not
pay attention to it.

665
00:29:05,660 --> 00:29:10,990
So in this case, it'll be V hat
is v, the visible component,

666
00:29:10,990 --> 00:29:12,320
plus 1 minus theta times o.

667
00:29:12,320 --> 00:29:15,280
And that's just repeating
what I showed you before.

668
00:29:15,280 --> 00:29:19,810
And that equals v minus
1 minus theta times tp,

669
00:29:19,810 --> 00:29:21,610
which is the sales tax.

670
00:29:21,610 --> 00:29:25,840
And now to the extent that
people pay attention to theta,

671
00:29:25,840 --> 00:29:30,370
the sales tax might
matter more or less.

672
00:29:30,370 --> 00:29:32,620
There might be other
opaque components

673
00:29:32,620 --> 00:29:35,082
that are unobservable
both to the experimenter

674
00:29:35,082 --> 00:29:36,040
but also to the person.

675
00:29:36,040 --> 00:29:37,310
They can abstract from that.

676
00:29:37,310 --> 00:29:39,727
So for now, they can essentially
say the less visible part

677
00:29:39,727 --> 00:29:43,090
is only the state tax.

678
00:29:43,090 --> 00:29:45,070
OK.

679
00:29:45,070 --> 00:29:46,120
Any questions on this?

680
00:29:51,110 --> 00:29:52,010
OK.

681
00:29:52,010 --> 00:29:54,320
So now, and this
is sort of like--

682
00:29:54,320 --> 00:29:58,010
so now, essentially,
what this gives us now,

683
00:29:58,010 --> 00:30:01,370
we can essentially try
to see if we have this--

684
00:30:03,950 --> 00:30:06,440
how can we identify
this parameter theta

685
00:30:06,440 --> 00:30:08,750
by making things very salient?

686
00:30:08,750 --> 00:30:13,510
And so the idea is now to
make the tax fully salient.

687
00:30:13,510 --> 00:30:15,783
Let me show you how this
looks like in practice.

688
00:30:15,783 --> 00:30:16,450
Here's the math.

689
00:30:16,450 --> 00:30:18,300
But I'll go back to
this in a second.

690
00:30:18,300 --> 00:30:22,130
Let me just show you what this
looks like in the experiment.

691
00:30:22,130 --> 00:30:24,760
So the idea in the
experiment is to say--

692
00:30:24,760 --> 00:30:29,200
you see on the lower part
here, you see a normal price.

693
00:30:29,200 --> 00:30:30,940
You see an original
price tag here.

694
00:30:30,940 --> 00:30:33,120
This is what the price tag
usually would look like.

695
00:30:33,120 --> 00:30:34,240
So this would be an item.

696
00:30:34,240 --> 00:30:37,360
I guess this is brushes.

697
00:30:37,360 --> 00:30:41,357
This item would
cost $5.79 usually.

698
00:30:41,357 --> 00:30:43,065
And then what they
did in the experiment,

699
00:30:43,065 --> 00:30:45,910
they essentially added a tag
where they said, now here,

700
00:30:45,910 --> 00:30:49,240
%5.79 plus sales tax is $6.22.

701
00:30:49,240 --> 00:30:51,190
And now they make
it entirely salient

702
00:30:51,190 --> 00:30:53,510
that there's a sales
tax, and you make

703
00:30:53,510 --> 00:30:56,070
the true price very salient.

704
00:30:56,070 --> 00:30:56,570
OK.

705
00:30:56,570 --> 00:30:57,945
And now the question
is if you do

706
00:30:57,945 --> 00:31:05,350
that, if we go from theta
being, for example, say 5.5,

707
00:31:05,350 --> 00:31:09,100
you only partially
observe the sales tax.

708
00:31:09,100 --> 00:31:12,970
If you go from that to making
the sales tax fully salient,

709
00:31:12,970 --> 00:31:15,010
what happens to people's demand?

710
00:31:15,010 --> 00:31:19,780
And can you infer from that
theta, what theta actually

711
00:31:19,780 --> 00:31:20,560
looks like?

712
00:31:20,560 --> 00:31:22,270
So let me go back to this.

713
00:31:22,270 --> 00:31:25,180
So essentially, what
we're trying to look at,

714
00:31:25,180 --> 00:31:28,480
we're trying to look
at a change in demand

715
00:31:28,480 --> 00:31:30,460
when theta falls to 0.

716
00:31:30,460 --> 00:31:31,750
Why does theta fall to 0?

717
00:31:31,750 --> 00:31:33,280
The assumption here
is that whatever

718
00:31:33,280 --> 00:31:35,770
is theta people have before,
so you might partially

719
00:31:35,770 --> 00:31:38,740
or even not at all--

720
00:31:38,740 --> 00:31:42,220
theta might be 0.5 or 1
or whatever for people.

721
00:31:42,220 --> 00:31:43,870
But now the
intervention that I just

722
00:31:43,870 --> 00:31:46,330
showed you brings
theta down to 0, right?

723
00:31:46,330 --> 00:31:48,520
Because now it's fully
salient to people

724
00:31:48,520 --> 00:31:50,200
what the sales tax is.

725
00:31:50,200 --> 00:31:52,420
And so the question
is now, can we

726
00:31:52,420 --> 00:31:54,670
have an expression for
the change in log demand

727
00:31:54,670 --> 00:32:00,880
when theta falls from whatever
it is before down to 0?

728
00:32:00,880 --> 00:32:05,260
So now what we have is
the change in log demand

729
00:32:05,260 --> 00:32:11,385
is the demand that
we had before.

730
00:32:11,385 --> 00:32:12,760
This is what I
showed you before.

731
00:32:12,760 --> 00:32:15,370
This was essentially demand
as a function of V. Remember

732
00:32:15,370 --> 00:32:16,620
what I showed you here before.

733
00:32:16,620 --> 00:32:17,412
Give me one second.

734
00:32:17,412 --> 00:32:18,270
I'll go back.

735
00:32:18,270 --> 00:32:20,910
So what I showed you here,
demand is a function of V hat.

736
00:32:20,910 --> 00:32:22,350
And V hat is this animal here.

737
00:32:22,350 --> 00:32:26,070
It's v minus 1 minus
theta times tp.

738
00:32:26,070 --> 00:32:29,190
Now, the question is, what
happens to demand when

739
00:32:29,190 --> 00:32:30,987
or log demand here--
forget about the log.

740
00:32:30,987 --> 00:32:32,070
It's not that interesting.

741
00:32:32,070 --> 00:32:33,570
That's just for math.

742
00:32:33,570 --> 00:32:36,830
So here's the demand as a
function of what it was before.

743
00:32:36,830 --> 00:32:39,420
And here's demand
when I now make

744
00:32:39,420 --> 00:32:41,010
things very salient, right?

745
00:32:41,010 --> 00:32:43,490
So now here in this
case, theta equals 0.

746
00:32:43,490 --> 00:32:46,945
And here, theta is
whatever it had before.

747
00:32:46,945 --> 00:32:48,570
Now when you have
this expression, what

748
00:32:48,570 --> 00:32:52,650
you can do is you can linearize
that, take the derivative

749
00:32:52,650 --> 00:32:56,195
and say, what is
this difference?

750
00:32:56,195 --> 00:32:58,320
Well, this difference is--
and you can do the math.

751
00:32:58,320 --> 00:33:02,640
It's essentially theta times tp.

752
00:33:02,640 --> 00:33:04,440
And then here, this
is the derivative

753
00:33:04,440 --> 00:33:05,790
with respect to theta.

754
00:33:05,790 --> 00:33:08,700
And what this gives
you essentially

755
00:33:08,700 --> 00:33:12,210
is that the change in log
demand is a function of theta,

756
00:33:12,210 --> 00:33:14,640
a function of tau,
which is like the tax,

757
00:33:14,640 --> 00:33:19,060
and eta, which is
the price elasticity.

758
00:33:19,060 --> 00:33:22,260
So essentially, that's to say
if you have now an intervention,

759
00:33:22,260 --> 00:33:25,470
we can look at, how does
demand or log demand change

760
00:33:25,470 --> 00:33:27,060
when we do this
intervention, assuming

761
00:33:27,060 --> 00:33:29,280
that this intervention
brings theta from whatever

762
00:33:29,280 --> 00:33:31,740
it was before down to 0?

763
00:33:31,740 --> 00:33:33,930
We can look at now
what happens to demand.

764
00:33:33,930 --> 00:33:35,550
That's delta log demand here.

765
00:33:35,550 --> 00:33:37,123
That's just a definition.

766
00:33:37,123 --> 00:33:38,040
And you can have that.

767
00:33:38,040 --> 00:33:42,900
That's an expression of how we
can relate that to theta to t,

768
00:33:42,900 --> 00:33:46,170
which is like the taxes--
it is at something like 7%--

769
00:33:46,170 --> 00:33:49,570
and then eta, which is the
price elasticity of demand.

770
00:33:49,570 --> 00:33:52,170
So that is to say, if we have
this change in demand that we

771
00:33:52,170 --> 00:33:54,630
observe from changing
theta down to 0

772
00:33:54,630 --> 00:33:58,740
and if we had the price
elasticity from just

773
00:33:58,740 --> 00:34:01,540
other variation in prices
that we see in the world

774
00:34:01,540 --> 00:34:04,740
and if you have t, which is
just how much the tax is,

775
00:34:04,740 --> 00:34:08,130
that would allow us
to identify theta.

776
00:34:08,130 --> 00:34:11,010
And that's what the
experiment is trying to do.

777
00:34:11,010 --> 00:34:13,560
So that implies essentially--
so we can just rearrange this.

778
00:34:13,560 --> 00:34:18,330
We can just say theta is, just
sort of rearranging things,

779
00:34:18,330 --> 00:34:22,860
minus the change in log demand
from this intervention divided

780
00:34:22,860 --> 00:34:26,389
by t times the price elasticity.

781
00:34:26,389 --> 00:34:28,699
OK, and that is just
to say essentially

782
00:34:28,699 --> 00:34:32,090
what we're trying to see is,
intuitively, that's just to say

783
00:34:32,090 --> 00:34:35,690
once we make things very salient
going from things not being

784
00:34:35,690 --> 00:34:38,449
salient to start with to
making things very salient,

785
00:34:38,449 --> 00:34:41,610
how much does
demand change there?

786
00:34:41,610 --> 00:34:48,270
And then we can
look at, well, for--

787
00:34:48,270 --> 00:34:52,480
in this case, I guess, this is
the taxes made very salient.

788
00:34:52,480 --> 00:34:54,219
And then we can
look at, well, how

789
00:34:54,219 --> 00:34:55,840
does this compare
to other changes

790
00:34:55,840 --> 00:34:58,870
in prices, assuming that
essentially people are paying

791
00:34:58,870 --> 00:35:00,580
attention to prices anyway?

792
00:35:00,580 --> 00:35:05,290
So if you look at things
becoming more expensive by 1%,

793
00:35:05,290 --> 00:35:08,470
how much are people changing
their behavior there?

794
00:35:08,470 --> 00:35:11,800
And now we have to take into
account that what's being

795
00:35:11,800 --> 00:35:13,750
made salient here is the tax t.

796
00:35:13,750 --> 00:35:16,480
And from backing out the
relative changes in demand

797
00:35:16,480 --> 00:35:19,810
for usual price changes
versus making things salient,

798
00:35:19,810 --> 00:35:22,750
we can now try to measure
theta as how much did they

799
00:35:22,750 --> 00:35:24,830
pay attention before.

800
00:35:24,830 --> 00:35:28,870
So one way to say this is for
example, if we make things very

801
00:35:28,870 --> 00:35:30,930
salient and the
change in log demand

802
00:35:30,930 --> 00:35:33,320
did not-- there's no
change in log demand,

803
00:35:33,320 --> 00:35:36,050
what does that imply for theta?

804
00:35:36,050 --> 00:35:37,300
What would we learn from that?

805
00:35:37,300 --> 00:35:39,105
So what was the change
here in demand at 0?

806
00:35:39,105 --> 00:35:40,730
People did not change
the demand at all

807
00:35:40,730 --> 00:35:42,760
when you make
things very salient

808
00:35:42,760 --> 00:35:44,980
versus what it was before.

809
00:35:44,980 --> 00:35:48,130
Or put differently, if
theta equals 0 here,

810
00:35:48,130 --> 00:35:52,270
then these two expressions are
the same, which essentially

811
00:35:52,270 --> 00:35:55,180
means that this thing here
is also going to be 0.

812
00:35:55,180 --> 00:36:01,750
So essentially, to the extent
that our intervention induces

813
00:36:01,750 --> 00:36:10,030
large changes in demand, that
must mean that essentially,

814
00:36:10,030 --> 00:36:14,010
people were very
attentive to start with.

815
00:36:14,010 --> 00:36:16,200
Essentially, if I'm making
the prices very salient

816
00:36:16,200 --> 00:36:20,310
and nothing happens, well,
that means people were already

817
00:36:20,310 --> 00:36:22,590
paying attention before.

818
00:36:22,590 --> 00:36:25,710
Now, in the maximum
case, if people

819
00:36:25,710 --> 00:36:29,760
were paying no attention
to prices before,

820
00:36:29,760 --> 00:36:31,120
what would happen then?

821
00:36:31,120 --> 00:36:33,870
Well, in that case, I
guess-- oops, sorry.

822
00:36:33,870 --> 00:36:36,660
In that case, theta would be 1.

823
00:36:36,660 --> 00:36:38,490
So essentially, it
would just be going

824
00:36:38,490 --> 00:36:41,730
from-- we would look at what's
the demand or change in demand

825
00:36:41,730 --> 00:36:46,380
from going v minus tp to
v, which is essentially

826
00:36:46,380 --> 00:36:48,900
just a change in tp.

827
00:36:48,900 --> 00:36:55,080
So that would be essentially
say if the tax was 7%, now

828
00:36:55,080 --> 00:36:57,030
we just have to look at
what is the equivalent

829
00:36:57,030 --> 00:37:01,380
effect of a 7% increase
in the price, which

830
00:37:01,380 --> 00:37:03,672
is exactly what we have here.

831
00:37:03,672 --> 00:37:05,130
So essentially,
just to summarize--

832
00:37:05,130 --> 00:37:07,560
and I think the math is, in some
sense, not that interesting.

833
00:37:07,560 --> 00:37:09,030
But essentially,
to summarize, it's

834
00:37:09,030 --> 00:37:12,510
like the larger our reaction
is to this intervention,

835
00:37:12,510 --> 00:37:14,490
the more inattentive
most people have

836
00:37:14,490 --> 00:37:19,410
been before because if they're
inattentive to start with,

837
00:37:19,410 --> 00:37:22,040
they react more to making
things more salient.

838
00:37:22,040 --> 00:37:26,180
So let me show you what
this looks like in practice.

839
00:37:26,180 --> 00:37:28,760
Again, the goal here is to
estimate a change in demand

840
00:37:28,760 --> 00:37:30,710
from making taxes fully salient.

841
00:37:30,710 --> 00:37:33,380
Again, here's the
original tag, and here's

842
00:37:33,380 --> 00:37:35,600
the experimental tag.

843
00:37:35,600 --> 00:37:38,060
It's very, very salient, very,
very clear, and very hard

844
00:37:38,060 --> 00:37:38,930
to miss for people.

845
00:37:38,930 --> 00:37:40,305
And the assumption
essentially is

846
00:37:40,305 --> 00:37:44,120
that people pay now full
attention to this new tag.

847
00:37:44,120 --> 00:37:45,920
They have a three-week
period in which--

848
00:37:45,920 --> 00:37:48,657
and this is an
experiment in California.

849
00:37:48,657 --> 00:37:50,240
They have a three-week
period in which

850
00:37:50,240 --> 00:37:53,700
they modify price
tags of certain items

851
00:37:53,700 --> 00:37:55,700
so that they go essentially
to different stores.

852
00:37:55,700 --> 00:37:56,690
They pick one store.

853
00:37:56,690 --> 00:37:59,060
And then essentially,
they modify the price tag

854
00:37:59,060 --> 00:38:02,090
only of the subset
of randomized goods.

855
00:38:02,090 --> 00:38:06,950
And they make the after-tax
price salient in addition

856
00:38:06,950 --> 00:38:08,630
to the pre-tax price, right?

857
00:38:08,630 --> 00:38:11,390
The pre-tax price is
already very salient here.

858
00:38:11,390 --> 00:38:16,073
In addition, they make the
after-tax price very salient.

859
00:38:16,073 --> 00:38:17,490
And now they're
going to do what's

860
00:38:17,490 --> 00:38:21,360
called a triple-diff estimate,
which is, in some sense, a very

861
00:38:21,360 --> 00:38:24,420
funky or fancy word just to
say what we're going to do

862
00:38:24,420 --> 00:38:29,280
is we compare the
sales or the demand

863
00:38:29,280 --> 00:38:32,080
during the treatment
periods to the following.

864
00:38:32,080 --> 00:38:33,360
So there's a treatment period.

865
00:38:33,360 --> 00:38:37,140
We compare how much of certain
goods have been treated.

866
00:38:37,140 --> 00:38:40,860
How much do the sales change
compared to the previous week

867
00:38:40,860 --> 00:38:43,002
sales for the same items?

868
00:38:43,002 --> 00:38:44,460
Second, we're going
to compare them

869
00:38:44,460 --> 00:38:47,940
to sales for items in which
the tax was not made salient.

870
00:38:47,940 --> 00:38:51,720
So there's some goods in the
store for which the tax was not

871
00:38:51,720 --> 00:38:52,950
made salient.

872
00:38:52,950 --> 00:38:56,140
And then finally, there's
going to be control stores.

873
00:38:56,140 --> 00:38:58,050
So we're going to
look at essentially,

874
00:38:58,050 --> 00:39:00,240
in the stores that
receive the treatment

875
00:39:00,240 --> 00:39:02,850
or in the store that receives
the treatment, what happens

876
00:39:02,850 --> 00:39:05,160
to the goods that receive
the treatment compared

877
00:39:05,160 --> 00:39:07,650
to the goods that don't
receive the treatment?

878
00:39:07,650 --> 00:39:11,940
We compare that to demand
before the treatment was

879
00:39:11,940 --> 00:39:15,235
enacted for both of
those types of goods.

880
00:39:15,235 --> 00:39:16,860
Essentially, what's
the relative change

881
00:39:16,860 --> 00:39:20,940
in demand for the treated versus
the non-treated items before

882
00:39:20,940 --> 00:39:21,555
versus after?

883
00:39:21,555 --> 00:39:22,930
So that's essentially
what people

884
00:39:22,930 --> 00:39:26,710
would call a diff in diff,
like difference in differences.

885
00:39:26,710 --> 00:39:28,980
And in addition to that,
they compare that difference

886
00:39:28,980 --> 00:39:31,770
in differences to the same
difference in differences

887
00:39:31,770 --> 00:39:32,715
in the control stores.

888
00:39:32,715 --> 00:39:36,010
So let me show you exactly
what this looks like.

889
00:39:36,010 --> 00:39:37,410
So this is a bit
of a messy table

890
00:39:37,410 --> 00:39:39,430
or a table with a
lot of information.

891
00:39:39,430 --> 00:39:41,380
But it's in fact quite simple.

892
00:39:41,380 --> 00:39:44,280
So what we have here is at
the top is a treatment store.

893
00:39:44,280 --> 00:39:46,970
And at the bottom, there's
some control stores.

894
00:39:46,970 --> 00:39:49,500
So in the control stores,
nothing has been changed.

895
00:39:49,500 --> 00:39:52,050
So there's no experiment.

896
00:39:52,050 --> 00:39:54,960
There's no additional tax
and so on being added.

897
00:39:54,960 --> 00:39:57,645
In the treatment stores, there's
going to be control categories

898
00:39:57,645 --> 00:39:59,640
and treated categories.

899
00:39:59,640 --> 00:40:01,890
Notice that they didn't fully
randomize all the goods.

900
00:40:01,890 --> 00:40:04,140
So it's not like if there's
one brush next to another,

901
00:40:04,140 --> 00:40:06,420
one brush has taxes versus not.

902
00:40:06,420 --> 00:40:09,090
Presumably, because they're
trying to minimize spillovers

903
00:40:09,090 --> 00:40:12,125
in some sense, if you only
make this tax salient for one

904
00:40:12,125 --> 00:40:13,500
good that's next
to another good,

905
00:40:13,500 --> 00:40:15,917
people would get very confused
because they're like, well,

906
00:40:15,917 --> 00:40:18,510
why does this one have
taxes made salient and then

907
00:40:18,510 --> 00:40:19,590
the other one not?

908
00:40:19,590 --> 00:40:22,200
So now this is across
categories where people maybe

909
00:40:22,200 --> 00:40:24,780
only pay attention to certain
goods in certain categories

910
00:40:24,780 --> 00:40:26,630
to start with.

911
00:40:26,630 --> 00:40:28,810
So what we're seeing
here is now here,

912
00:40:28,810 --> 00:40:31,885
these are the control categories
and the treated categories.

913
00:40:31,885 --> 00:40:34,510
And we can look at what's called
the difference in differences.

914
00:40:34,510 --> 00:40:37,200
We can look at the change in
demand in the treatment stores.

915
00:40:37,200 --> 00:40:38,440
What we're going to
look at is essentially

916
00:40:38,440 --> 00:40:40,540
the change in demand in
the control categories

917
00:40:40,540 --> 00:40:42,440
and the treatment categories.

918
00:40:42,440 --> 00:40:46,150
So what we see here
is we can compare

919
00:40:46,150 --> 00:40:48,670
how much is sold in the
experimental period, which

920
00:40:48,670 --> 00:40:52,270
is this row here,
compared to our baseline.

921
00:40:52,270 --> 00:40:55,210
What we see is essentially
demand in the control

922
00:40:55,210 --> 00:40:57,850
categories was going up
a little bit by 0.84.

923
00:40:57,850 --> 00:40:59,590
That's just because
of seasonalities.

924
00:40:59,590 --> 00:41:01,840
Maybe people were just
buying more anyway

925
00:41:01,840 --> 00:41:03,250
for whatever reason.

926
00:41:03,250 --> 00:41:06,100
And in the treated categories
over the same period,

927
00:41:06,100 --> 00:41:08,260
comparing essentially
the experimental periods

928
00:41:08,260 --> 00:41:12,920
to the baseline period,
demand went down.

929
00:41:12,920 --> 00:41:14,800
So now the difference
in differences estimate

930
00:41:14,800 --> 00:41:17,620
is essentially the change in
demand in the treatment stores,

931
00:41:17,620 --> 00:41:20,153
comparing the treated categories
compared to the control

932
00:41:20,153 --> 00:41:21,820
categories, which is
just the difference

933
00:41:21,820 --> 00:41:24,170
between this column
and this column,

934
00:41:24,170 --> 00:41:27,570
which is minus 2.14 units.

935
00:41:27,570 --> 00:41:28,300
OK.

936
00:41:28,300 --> 00:41:31,150
So that's essentially,
how much did the demand

937
00:41:31,150 --> 00:41:34,750
change before versus after in
the treated categories compared

938
00:41:34,750 --> 00:41:38,330
to the same change in
the control categories?

939
00:41:38,330 --> 00:41:41,090
So that's essentially
a difference

940
00:41:41,090 --> 00:41:43,240
in differences estimator.

941
00:41:43,240 --> 00:41:45,560
Now, they also have
the control stores.

942
00:41:45,560 --> 00:41:48,110
Why would we want the
control stores, as well?

943
00:41:48,110 --> 00:41:50,510
Why is that helpful to have?

944
00:41:50,510 --> 00:41:53,930
Suppose that improved sunscreen,
and now it's getting warmer.

945
00:41:53,930 --> 00:41:57,410
It's really hot, and everybody
was out buying sunscreen.

946
00:41:57,410 --> 00:41:59,540
Then you see essentially
people's demand

947
00:41:59,540 --> 00:42:00,645
in the control categories.

948
00:42:00,645 --> 00:42:02,270
The treatment category
goes up by a lot

949
00:42:02,270 --> 00:42:04,070
or goes down by a lot
for whatever reason.

950
00:42:04,070 --> 00:42:07,070
Suppose it's umbrellas, and
one week was really rainy,

951
00:42:07,070 --> 00:42:08,780
and the other one was not rainy.

952
00:42:08,780 --> 00:42:12,140
And we see now essentially
seasonal or other differences

953
00:42:12,140 --> 00:42:15,297
over time in some of the
categories that would,

954
00:42:15,297 --> 00:42:16,880
even if it's randomized,
it could just

955
00:42:16,880 --> 00:42:18,480
happen to be the case.

956
00:42:18,480 --> 00:42:21,230
So that would not have to
do anything with a sales tax

957
00:42:21,230 --> 00:42:21,890
being salient.

958
00:42:21,890 --> 00:42:23,508
That's just by chance.

959
00:42:23,508 --> 00:42:25,550
So now it's very nice to
have some control stores

960
00:42:25,550 --> 00:42:27,410
where we can look at like the
same difference in difference

961
00:42:27,410 --> 00:42:30,060
estimates and some of the
control stores, in some sense,

962
00:42:30,060 --> 00:42:31,193
like a placebo.

963
00:42:31,193 --> 00:42:32,360
They can do the same things.

964
00:42:32,360 --> 00:42:34,860
There's going to be the control
categories and the treatment

965
00:42:34,860 --> 00:42:35,960
categories.

966
00:42:35,960 --> 00:42:38,150
Notice, to be very clear,
in the control stores,

967
00:42:38,150 --> 00:42:42,190
the treated categories do not
receive any actual treatment.

968
00:42:42,190 --> 00:42:44,330
So we can do the same
difference in differences.

969
00:42:44,330 --> 00:42:47,900
And what we see here is now
that there's essentially

970
00:42:47,900 --> 00:42:48,750
no difference here.

971
00:42:48,750 --> 00:42:51,110
So the demand in the
control categories

972
00:42:51,110 --> 00:42:54,403
went up a little bit for
whatever reason in the treated.

973
00:42:54,403 --> 00:42:56,820
And this is a little bit smaller
than in this other store.

974
00:42:56,820 --> 00:42:58,950
This difference is not
statistically significant.

975
00:42:58,950 --> 00:43:01,200
But anyway, the control
categories go up a little bit.

976
00:43:01,200 --> 00:43:03,433
The treated categories
go also up a little bit.

977
00:43:03,433 --> 00:43:05,600
When you do this difference
in differences estimate,

978
00:43:05,600 --> 00:43:07,453
you find essentially
no difference.

979
00:43:07,453 --> 00:43:08,870
In the control
categories, there's

980
00:43:08,870 --> 00:43:11,360
no change in the
treated categories

981
00:43:11,360 --> 00:43:13,610
compared to the baseline
compared to the control

982
00:43:13,610 --> 00:43:14,200
categories.

983
00:43:14,200 --> 00:43:16,820
So difference in difference
estimate in the control

984
00:43:16,820 --> 00:43:18,722
category is essentially 0.

985
00:43:18,722 --> 00:43:20,930
And now what you can do is
essentially the difference

986
00:43:20,930 --> 00:43:22,190
in differences estimate.

987
00:43:22,190 --> 00:43:22,940
You can compare.

988
00:43:22,940 --> 00:43:25,130
You can subtract
this number here

989
00:43:25,130 --> 00:43:27,170
minus this number
here, which gives you

990
00:43:27,170 --> 00:43:31,100
the triple-diff estimate,
which essentially

991
00:43:31,100 --> 00:43:34,070
is the relative change of
treatment versus control

992
00:43:34,070 --> 00:43:37,580
before and after and treatment
versus control stores.

993
00:43:37,580 --> 00:43:41,970
It's a little bit lots of
differences in your head.

994
00:43:41,970 --> 00:43:45,570
But I hope this comparison
from the table makes it clear.

995
00:43:45,570 --> 00:43:48,728
Any questions on the
estimate or the procedure

996
00:43:48,728 --> 00:43:49,520
that they're doing?

997
00:43:58,750 --> 00:43:59,380
OK.

998
00:43:59,380 --> 00:44:02,200
So what we see here now
is essentially now making

999
00:44:02,200 --> 00:44:05,290
these taxes salient in
the treated categories

1000
00:44:05,290 --> 00:44:09,010
in the treatment stores
reduces demand, right?

1001
00:44:09,010 --> 00:44:10,670
Because essentially,
demand goes down.

1002
00:44:10,670 --> 00:44:11,860
Why does the demand go down?

1003
00:44:11,860 --> 00:44:16,678
Well, the perceived price that
people face has now gone up.

1004
00:44:16,678 --> 00:44:18,220
And now they buy
less of these goods,

1005
00:44:18,220 --> 00:44:21,280
presumably because before,
they were not paying attention.

1006
00:44:21,280 --> 00:44:22,570
Now the price goes up.

1007
00:44:22,570 --> 00:44:25,570
And now, essentially,
demand falls, as it should,

1008
00:44:25,570 --> 00:44:29,650
because now things
are more expensive.

1009
00:44:29,650 --> 00:44:32,690
And this is here the result. The
average quantity sold decreases

1010
00:44:32,690 --> 00:44:35,600
by 2.2 units-- this is
units of the goods--

1011
00:44:35,600 --> 00:44:37,610
relative to baseline
level of 25.

1012
00:44:37,610 --> 00:44:41,240
That's an 8.8% decline.

1013
00:44:41,240 --> 00:44:42,965
And now we can use
our own formula

1014
00:44:42,965 --> 00:44:46,700
and compute the
degree of inattention.

1015
00:44:46,700 --> 00:44:54,440
Essentially, now we can say, OK,
now we know what the tax was.

1016
00:44:54,440 --> 00:44:57,850
The tax was 7.375%.

1017
00:44:57,850 --> 00:45:02,390
We can estimate the price
elasticity in some other ways.

1018
00:45:02,390 --> 00:45:05,380
Essentially, we can say,
what happens if prices go up

1019
00:45:05,380 --> 00:45:07,728
by 1% for other reasons?

1020
00:45:07,728 --> 00:45:10,270
What if the salient component
that's already salient goes up?

1021
00:45:10,270 --> 00:45:13,120
This is week-by-week
variation across stores.

1022
00:45:13,120 --> 00:45:16,480
So if the prices go up,
what's the price elasticity?

1023
00:45:16,480 --> 00:45:18,370
And the estimate of
those price elasticity

1024
00:45:18,370 --> 00:45:21,100
is also given from
other variation.

1025
00:45:21,100 --> 00:45:22,660
That's 1.59.

1026
00:45:22,660 --> 00:45:24,598
And now we can essentially
back out theta.

1027
00:45:24,598 --> 00:45:26,140
Essentially, what
we're trying to ask

1028
00:45:26,140 --> 00:45:34,510
is we have a change in 2.2 units
that is making prices salient.

1029
00:45:34,510 --> 00:45:36,880
Now the question is, well,
how large of a change

1030
00:45:36,880 --> 00:45:42,220
is it relative to a change in
prices of 1%, 2%, and so on?

1031
00:45:42,220 --> 00:45:47,440
We know that essentially the
price elasticity is 1.59.

1032
00:45:47,440 --> 00:45:50,780
So we can essentially
just back out theta.

1033
00:45:50,780 --> 00:45:52,360
And so what we get
from this formula

1034
00:45:52,360 --> 00:45:58,150
now is that theta is about 0.75,
which says that consumers react

1035
00:45:58,150 --> 00:46:01,480
to price changes due to sales
tax changes only a quarter as

1036
00:46:01,480 --> 00:46:03,490
much as to other price changes.

1037
00:46:03,490 --> 00:46:08,200
So essentially, consumers are
very inattentive, in fact,

1038
00:46:08,200 --> 00:46:12,080
to sales taxes, to
sales tax changes.

1039
00:46:12,080 --> 00:46:13,840
In fact, they only
perceive a quarter

1040
00:46:13,840 --> 00:46:19,280
as much as any other price
changes in the world.

1041
00:46:19,280 --> 00:46:21,730
And that's really saying,
look, people really

1042
00:46:21,730 --> 00:46:23,980
seem to be missing these
things that are not salient.

1043
00:46:23,980 --> 00:46:27,080
Now if you make them salient,
people change their behavior.

1044
00:46:27,080 --> 00:46:31,000
Therefore, they must have been
inattentive to start with.

1045
00:46:31,000 --> 00:46:33,805
In addition to that
kind of evidence,

1046
00:46:33,805 --> 00:46:35,500
Chetty et al. also
have some evidence

1047
00:46:35,500 --> 00:46:38,770
on using non-experimental
variation.

1048
00:46:38,770 --> 00:46:40,810
And in some sense, you
could say the evidence

1049
00:46:40,810 --> 00:46:43,602
that I just showed you from
the store is a little bit

1050
00:46:43,602 --> 00:46:46,060
weird because, in some sense,
you see these different tags.

1051
00:46:46,060 --> 00:46:48,640
And maybe the tags
raise attention

1052
00:46:48,640 --> 00:46:50,740
to the price even
regardless of the tax.

1053
00:46:50,740 --> 00:46:53,150
And it's a little bit
weird in some ways.

1054
00:46:53,150 --> 00:46:55,720
So it's quite nice to have
some complementary evidence

1055
00:46:55,720 --> 00:47:03,140
of a non-experimental
panel deliberation.

1056
00:47:03,140 --> 00:47:07,810
And so what they do is they
look at beer consumption.

1057
00:47:07,810 --> 00:47:12,640
And it turns out beer, tobacco,
and other types of goods

1058
00:47:12,640 --> 00:47:16,600
have two types of
taxes levied on them.

1059
00:47:16,600 --> 00:47:20,020
They have an excise tax, which
is included in the price.

1060
00:47:20,020 --> 00:47:23,773
This is highly salient
during choice the process

1061
00:47:23,773 --> 00:47:25,690
because essentially it's
in the price already.

1062
00:47:25,690 --> 00:47:27,997
So if taxes go up,
essentially, the price goes up.

1063
00:47:27,997 --> 00:47:29,830
And people will notice
if they pay attention

1064
00:47:29,830 --> 00:47:32,170
or since they pay attention
to prices, since they

1065
00:47:32,170 --> 00:47:36,280
like the price tag already.

1066
00:47:36,280 --> 00:47:37,660
And then there's a sales tax.

1067
00:47:37,660 --> 00:47:39,970
This is the tax that we were
just talking about before.

1068
00:47:39,970 --> 00:47:42,460
That's essentially very opaque
during the choice process,

1069
00:47:42,460 --> 00:47:43,810
as we just showed.

1070
00:47:43,810 --> 00:47:46,570
And now we can look at
variation across states

1071
00:47:46,570 --> 00:47:52,300
and over time of changes in
excise taxes and sales taxes.

1072
00:47:52,300 --> 00:47:54,280
And we can look at,
how much do people

1073
00:47:54,280 --> 00:47:57,700
react to excise taxes
relative to how much people

1074
00:47:57,700 --> 00:47:59,440
react to sales taxes.

1075
00:47:59,440 --> 00:48:01,450
And that relative reaction
will tell us, again,

1076
00:48:01,450 --> 00:48:04,030
something about theta.

1077
00:48:04,030 --> 00:48:07,210
And what Chetty et al. finds
in here is a table for that.

1078
00:48:07,210 --> 00:48:10,690
Essentially, when you look
at changes in excise taxes,

1079
00:48:10,690 --> 00:48:13,210
this tax is very salient.

1080
00:48:13,210 --> 00:48:18,850
The price elasticity
is above 0.8 or 0.9.

1081
00:48:18,850 --> 00:48:22,600
So essentially, when
the price goes up by 1%,

1082
00:48:22,600 --> 00:48:26,680
people react by about almost 1%.

1083
00:48:26,680 --> 00:48:30,400
In contrast, there's some
differences in estimates.

1084
00:48:30,400 --> 00:48:32,590
But once you control
for things properly,

1085
00:48:32,590 --> 00:48:35,560
people essentially do not
react to the sales tax at all.

1086
00:48:35,560 --> 00:48:39,520
Maybe a little bit, but overall,
the number's very, very low.

1087
00:48:39,520 --> 00:48:42,180
And now, essentially, the ratio
of these types of reaction

1088
00:48:42,180 --> 00:48:45,160
tells you how
attentive people are.

1089
00:48:45,160 --> 00:48:47,900
Essentially, if you
divide, say, for example--

1090
00:48:47,900 --> 00:48:53,427
if you took column number three,
if you divide 0.003 by 0.86,

1091
00:48:53,427 --> 00:48:55,510
and you have to do a little
bit of additional math

1092
00:48:55,510 --> 00:48:57,550
that's uninteresting--
you essentially

1093
00:48:57,550 --> 00:49:00,430
will get a result that,
essentially, people do not

1094
00:49:00,430 --> 00:49:03,610
pay attention to sales taxes.

1095
00:49:03,610 --> 00:49:06,490
1 minus theta is
eventually close to 0.

1096
00:49:06,490 --> 00:49:09,400
Put differently, the degree
of inattention is almost 1.

1097
00:49:09,400 --> 00:49:11,170
Theta is almost 1.

1098
00:49:11,170 --> 00:49:14,620
So essentially, people
do not pay attention

1099
00:49:14,620 --> 00:49:18,190
to these types of taxes.

1100
00:49:18,190 --> 00:49:20,740
Inattention is almost complete.

1101
00:49:20,740 --> 00:49:23,560
So there's substantial
consumer inattention

1102
00:49:23,560 --> 00:49:26,300
to non-transparent taxes.

1103
00:49:26,300 --> 00:49:29,760
So what I've shown you
so far is, essentially,

1104
00:49:29,760 --> 00:49:31,300
the existence of inattention.

1105
00:49:31,300 --> 00:49:33,550
It doesn't say why
people are inattentive.

1106
00:49:33,550 --> 00:49:37,040
It could be people just can't
do the math in their head.

1107
00:49:37,040 --> 00:49:39,040
It could be people don't
want to bother with it.

1108
00:49:39,040 --> 00:49:41,457
They're sort of saying, well,
it's kind of a small change.

1109
00:49:41,457 --> 00:49:42,520
So why do we really care?

1110
00:49:42,520 --> 00:49:45,607
It could be that people who
buy alcohol half of the time

1111
00:49:45,607 --> 00:49:47,440
are drunk, and they
don't pay any attention.

1112
00:49:47,440 --> 00:49:54,020
There's no modeling in any way
of, why are people doing this?

1113
00:49:54,020 --> 00:49:57,580
This is just saying this shows
the existence of inattention

1114
00:49:57,580 --> 00:50:00,070
in those kinds of
consumer choices.

1115
00:50:00,070 --> 00:50:02,020
It doesn't say why
people are inattentive.

1116
00:50:02,020 --> 00:50:04,120
And you might say,
well, 7% of a tax

1117
00:50:04,120 --> 00:50:05,770
is kind of a small distortion.

1118
00:50:05,770 --> 00:50:10,460
So it might be actually
optimal to miss that,

1119
00:50:10,460 --> 00:50:12,712
which is a reasonable approach.

1120
00:50:12,712 --> 00:50:14,170
It is probably
quite a bit of money

1121
00:50:14,170 --> 00:50:15,920
and quite a bit of
distortion if you think

1122
00:50:15,920 --> 00:50:17,740
about it and you actually--

1123
00:50:17,740 --> 00:50:19,210
7% is not nothing.

1124
00:50:19,210 --> 00:50:21,363
But if you have a
lot of money anyway,

1125
00:50:21,363 --> 00:50:23,780
you might as well just not pay
any attention to that stuff

1126
00:50:23,780 --> 00:50:27,700
and if that helps you pay
attention to other things.

1127
00:50:27,700 --> 00:50:30,100
Notice that even
then, that shows

1128
00:50:30,100 --> 00:50:31,960
the existence of inattention.

1129
00:50:31,960 --> 00:50:33,460
It's just a bit of
a question, like,

1130
00:50:33,460 --> 00:50:34,900
how important is inattention?

1131
00:50:34,900 --> 00:50:37,500
And this actually gets
me to the next point,

1132
00:50:37,500 --> 00:50:39,250
which is a question
of, why should we care

1133
00:50:39,250 --> 00:50:40,810
about inattention to taxes?

1134
00:50:40,810 --> 00:50:44,120
Why is that important or
interesting potentially?

1135
00:50:44,120 --> 00:50:46,205
Question exactly is,
who's paying attention?

1136
00:50:46,205 --> 00:50:47,330
Who's not paying attention?

1137
00:50:47,330 --> 00:50:49,010
And then who's
sort of distorting

1138
00:50:49,010 --> 00:50:50,280
their behavior potentially?

1139
00:50:50,280 --> 00:50:53,600
There's lots of interesting
issues about heterogeneity

1140
00:50:53,600 --> 00:50:56,150
and who's paying attention.

1141
00:50:56,150 --> 00:50:58,610
Let me talk about
this in a second.

1142
00:50:58,610 --> 00:51:00,680
Let me ask differently.

1143
00:51:00,680 --> 00:51:04,070
Why is it that some taxes, the
excise taxes, are very salient?

1144
00:51:04,070 --> 00:51:06,860
And why is it that other
taxes are not salient?

1145
00:51:06,860 --> 00:51:08,820
What do you think about that?

1146
00:51:08,820 --> 00:51:09,680
Is that an accident?

1147
00:51:09,680 --> 00:51:12,140
Or why is it that the taxes
for alcohol and tobacco,

1148
00:51:12,140 --> 00:51:13,910
et cetera, are very salient?

1149
00:51:13,910 --> 00:51:16,370
And why is it that other
taxes for shampoo or

1150
00:51:16,370 --> 00:51:19,580
whatever are
actually not salient?

1151
00:51:19,580 --> 00:51:23,055
There's two simple objectives
if you took a public economics

1152
00:51:23,055 --> 00:51:24,890
class, if you took
John Gruber's class,

1153
00:51:24,890 --> 00:51:28,070
but also if you
took 1401 or 1403.

1154
00:51:28,070 --> 00:51:31,910
The reason why the
government levies taxes is--

1155
00:51:31,910 --> 00:51:33,830
one big reason is to
generate revenue, right?

1156
00:51:33,830 --> 00:51:34,955
The government needs money.

1157
00:51:34,955 --> 00:51:36,690
Somehow, we need to tax people.

1158
00:51:36,690 --> 00:51:39,770
And the issue, of course,
is that if you levy taxes

1159
00:51:39,770 --> 00:51:42,470
on some goods
versus others, that

1160
00:51:42,470 --> 00:51:44,600
can distort consumer choices.

1161
00:51:44,600 --> 00:51:47,960
And there's essentially
some theorems

1162
00:51:47,960 --> 00:51:54,260
and some considerations about
which goods should you tax.

1163
00:51:54,260 --> 00:51:56,480
And ideally, you would
tax goods that people

1164
00:51:56,480 --> 00:51:58,490
are consuming anyway.

1165
00:51:58,490 --> 00:52:00,410
Essentially, what you
want is people make--

1166
00:52:00,410 --> 00:52:02,810
so essentially, when you
have taxes on some goods

1167
00:52:02,810 --> 00:52:05,450
but not on others, if
people have different price

1168
00:52:05,450 --> 00:52:07,490
elasticities on some
goods versus others

1169
00:52:07,490 --> 00:52:11,840
and if you increase one good or
all goods by a certain amount,

1170
00:52:11,840 --> 00:52:14,660
if people have different
price elasticities,

1171
00:52:14,660 --> 00:52:17,290
then you might distort
their behavior.

1172
00:52:17,290 --> 00:52:20,440
So if you increase and you
start taxing one thing and not

1173
00:52:20,440 --> 00:52:23,600
another, people might
shift to the other good.

1174
00:52:23,600 --> 00:52:26,690
And we don't want that because,
presumably, people are already

1175
00:52:26,690 --> 00:52:29,090
optimizing to start with.

1176
00:52:29,090 --> 00:52:31,370
Now, if people are
inelastic to certain goods,

1177
00:52:31,370 --> 00:52:33,620
if people need
certain goods anyway

1178
00:52:33,620 --> 00:52:34,970
and they buy them anyway--

1179
00:52:34,970 --> 00:52:37,220
and the last thing, essentially,
there's public goods.

1180
00:52:37,220 --> 00:52:41,000
Public economics would tell you
or some theories would tell you

1181
00:52:41,000 --> 00:52:44,030
you should tax these goods more.

1182
00:52:44,030 --> 00:52:47,450
Now, distortion, when people
are-- so the distortions

1183
00:52:47,450 --> 00:52:50,810
are lower if people don't
fully react to taxes.

1184
00:52:50,810 --> 00:52:54,030
That could be if, for example,
you started taxing bread,

1185
00:52:54,030 --> 00:52:57,890
for example, while everybody--
or certain goods like potatoes,

1186
00:52:57,890 --> 00:53:00,470
where everybody needs to
buy these goods anyway

1187
00:53:00,470 --> 00:53:02,360
because people need to eat--

1188
00:53:02,360 --> 00:53:06,620
well, then, there
would be very low--

1189
00:53:09,140 --> 00:53:12,430
these goods would
be very inelastic.

1190
00:53:12,430 --> 00:53:14,180
And that's because
they need those things.

1191
00:53:14,180 --> 00:53:15,740
Not because they're
inattentive, but just

1192
00:53:15,740 --> 00:53:17,365
because they need
these types of goods.

1193
00:53:17,365 --> 00:53:21,830
Now, there's obviously
issues with poverty.

1194
00:53:21,830 --> 00:53:24,320
We don't want to tax
the poor too much.

1195
00:53:24,320 --> 00:53:27,630
But in general, they want to
tax things that are inelastic.

1196
00:53:27,630 --> 00:53:30,940
And so in a way, if people
are inattentive to the taxes,

1197
00:53:30,940 --> 00:53:32,690
in a way, that's great
because now they're

1198
00:53:32,690 --> 00:53:33,970
not distorting their behavior.

1199
00:53:33,970 --> 00:53:36,360
And we can still
get money from them.

1200
00:53:36,360 --> 00:53:40,400
So in some sense, actually,
some of the inattention

1201
00:53:40,400 --> 00:53:42,380
for that purpose
might be actually good

1202
00:53:42,380 --> 00:53:45,620
because now people are not
distorting their behavior

1203
00:53:45,620 --> 00:53:49,310
by looking at those prices.

1204
00:53:49,310 --> 00:53:51,620
So being inattentive
in some ways is good.

1205
00:53:51,620 --> 00:53:54,720
But exactly as Maya was
saying, for some goods--

1206
00:53:54,720 --> 00:53:58,580
and these are often labeled
as sin goods or the like,

1207
00:53:58,580 --> 00:54:01,500
where there's either
externalities or internalities.

1208
00:54:01,500 --> 00:54:04,820
For example, if you think
about tobacco or alcohol,

1209
00:54:04,820 --> 00:54:05,870
there are externalities.

1210
00:54:05,870 --> 00:54:09,590
Tobacco, for example, secondhand
smoking is bad for others.

1211
00:54:09,590 --> 00:54:11,750
For alcohol, there's
often externalities

1212
00:54:11,750 --> 00:54:15,390
in terms of drunk driving
or violence or the like.

1213
00:54:15,390 --> 00:54:18,350
So the government wants to
increase the price of alcohol

1214
00:54:18,350 --> 00:54:22,430
because when people are making
choices of how much alcohol

1215
00:54:22,430 --> 00:54:25,190
to buy, they might not take into
account the effects of alcohol

1216
00:54:25,190 --> 00:54:26,420
on others.

1217
00:54:26,420 --> 00:54:29,570
By internalities, I mean
people might not necessarily

1218
00:54:29,570 --> 00:54:33,620
take into account the
effect of alcohol or tobacco

1219
00:54:33,620 --> 00:54:34,858
on themselves.

1220
00:54:34,858 --> 00:54:36,400
This is what we
discussed previously.

1221
00:54:36,400 --> 00:54:38,240
This could be because
of present bias.

1222
00:54:38,240 --> 00:54:41,900
This could be because of
biased beliefs or the like.

1223
00:54:41,900 --> 00:54:44,552
Essentially, people might
hurt their future selves.

1224
00:54:44,552 --> 00:54:46,010
And the government
might say, well,

1225
00:54:46,010 --> 00:54:49,070
let's increase taxes
because their future sales--

1226
00:54:49,070 --> 00:54:50,510
people might like
that there might

1227
00:54:50,510 --> 00:54:52,940
be a way of helping
people, for instance,

1228
00:54:52,940 --> 00:54:55,080
deal with self control problems.

1229
00:54:55,080 --> 00:54:57,980
So here, the
government explicitly

1230
00:54:57,980 --> 00:55:01,370
wants consumers to
react to the taxes.

1231
00:55:01,370 --> 00:55:03,188
The point is not to
make money from people.

1232
00:55:03,188 --> 00:55:04,730
And in fact, sometimes
you're worried

1233
00:55:04,730 --> 00:55:07,200
about making too much
money from people.

1234
00:55:07,200 --> 00:55:10,040
The point here is really
to change behavior

1235
00:55:10,040 --> 00:55:13,790
to reduce the behaviors
that are being taxed.

1236
00:55:13,790 --> 00:55:16,650
So you want to make such
taxes particularly salient.

1237
00:55:16,650 --> 00:55:17,150
Ready?

1238
00:55:17,150 --> 00:55:22,070
Now, look at the taxes on
alcohol versus the taxes on--

1239
00:55:22,070 --> 00:55:24,460
some of the excise taxes
versus the sales taxes.

1240
00:55:24,460 --> 00:55:27,710
The excise taxes are
precisely salient because we

1241
00:55:27,710 --> 00:55:31,828
want people to react to it
while the sales taxes are not.

1242
00:55:31,828 --> 00:55:33,620
Of course, there's
variation across places.

1243
00:55:33,620 --> 00:55:36,320
But it's not an accident that
we see this in the world the way

1244
00:55:36,320 --> 00:55:37,310
that is.

1245
00:55:37,310 --> 00:55:39,980
Now, one thing that's
quite interesting now is--

1246
00:55:39,980 --> 00:55:43,070
and I think José or
[INAUDIBLE] was saying that.

1247
00:55:43,070 --> 00:55:45,660
It's like when consumers
are heterogeneous,

1248
00:55:45,660 --> 00:55:47,850
then there's a lot of
quite interesting issues.

1249
00:55:47,850 --> 00:55:50,850
For example, if you have
the poor paying more or less

1250
00:55:50,850 --> 00:55:53,700
attention compared to the rich
or if only a subset of people

1251
00:55:53,700 --> 00:55:56,970
pay attention to those
goods, then in a way,

1252
00:55:56,970 --> 00:56:00,090
you would find that the average
elasticity is still quite low.

1253
00:56:00,090 --> 00:56:02,400
But some people are actually
distorting their behavior

1254
00:56:02,400 --> 00:56:04,860
quite a bit, and others do not.

1255
00:56:04,860 --> 00:56:11,010
And that gets relatively
quickly quite complicated.

1256
00:56:11,010 --> 00:56:13,063
And you could then
get into situations

1257
00:56:13,063 --> 00:56:15,480
where you make some people a
lot worse off and some people

1258
00:56:15,480 --> 00:56:20,150
better off when making
something salient versus not.

1259
00:56:20,150 --> 00:56:21,360
Any questions on all of this?

1260
00:56:28,460 --> 00:56:29,360
OK.

1261
00:56:29,360 --> 00:56:32,058
So then we get to
the next question.

1262
00:56:32,058 --> 00:56:34,100
This is getting a little
bit at the question that

1263
00:56:34,100 --> 00:56:37,340
was asked previously about,
can inattention or attention

1264
00:56:37,340 --> 00:56:39,200
have large effects?

1265
00:56:39,200 --> 00:56:44,300
And so on the one hand,
you might say, well,

1266
00:56:44,300 --> 00:56:46,550
people's choices-- for
example, consumption patterns--

1267
00:56:46,550 --> 00:56:49,280
are distorted due to
limited attention.

1268
00:56:49,280 --> 00:56:53,510
But you might ask, well,
if it's really true

1269
00:56:53,510 --> 00:56:56,090
that people's consumption
patterns have really distorted

1270
00:56:56,090 --> 00:56:59,330
that much, if I'm so much worse
off by not paying attention

1271
00:56:59,330 --> 00:57:01,910
to things, well, eventually,
I'm going to notice this.

1272
00:57:01,910 --> 00:57:03,800
Somebody is going to tell me.

1273
00:57:03,800 --> 00:57:09,320
Or in a way, I notice
myself eventually

1274
00:57:09,320 --> 00:57:11,810
that I'm doing things really
worse compared to my friends.

1275
00:57:11,810 --> 00:57:16,460
And eventually, if things
are really distorted a lot,

1276
00:57:16,460 --> 00:57:19,290
I might pay attention to it.

1277
00:57:19,290 --> 00:57:23,540
So the underlying idea is
that, potentially, people

1278
00:57:23,540 --> 00:57:26,660
are actually inattentive
in the sense of saying,

1279
00:57:26,660 --> 00:57:29,630
look, people's attention
is limited in some ways.

1280
00:57:29,630 --> 00:57:35,060
But it's optimal for
people to pay attention

1281
00:57:35,060 --> 00:57:36,770
to things that are
really important.

1282
00:57:36,770 --> 00:57:38,187
And there, the
underlying question

1283
00:57:38,187 --> 00:57:39,560
is, what is salient to people?

1284
00:57:39,560 --> 00:57:41,887
And how do people
decide what to focus on?

1285
00:57:41,887 --> 00:57:43,970
And won't people pay
attention to important things

1286
00:57:43,970 --> 00:57:47,473
anyway if they have to?

1287
00:57:47,473 --> 00:57:48,890
And so then the
question is, well,

1288
00:57:48,890 --> 00:57:50,750
how is it possible
that inattention could

1289
00:57:50,750 --> 00:57:52,780
have large effects anyway?

1290
00:57:52,780 --> 00:57:55,790
And so the next paper is
a good example of that.

1291
00:57:55,790 --> 00:57:58,820
And here's a very intuitive
example of why that might be.

1292
00:57:58,820 --> 00:58:00,570
So consider the
following situation.

1293
00:58:00,570 --> 00:58:02,960
You have been getting headaches.

1294
00:58:02,960 --> 00:58:04,550
You go to the doctor.

1295
00:58:04,550 --> 00:58:07,010
And the doctor asks you
whether it gets worse

1296
00:58:07,010 --> 00:58:08,900
after eating certain foods.

1297
00:58:08,900 --> 00:58:10,860
And so what are
you going to say?

1298
00:58:10,860 --> 00:58:15,590
Well, the answer will probably
be like, I don't know.

1299
00:58:15,590 --> 00:58:17,060
And why is that?

1300
00:58:17,060 --> 00:58:21,481
Why might you just not even know
the answer to that question?

1301
00:58:21,481 --> 00:58:24,310
An underlying
theory in your mind

1302
00:58:24,310 --> 00:58:26,500
of why you might be
getting headaches.

1303
00:58:26,500 --> 00:58:29,350
And you might just
not have a theory

1304
00:58:29,350 --> 00:58:31,340
that it's related to
your food consumption.

1305
00:58:31,340 --> 00:58:33,950
So while you have lots
of data in front of you--

1306
00:58:33,950 --> 00:58:36,160
you eat every day, you
get headaches on some days

1307
00:58:36,160 --> 00:58:37,900
and not on others-- you could
surely pay attention to it

1308
00:58:37,900 --> 00:58:38,860
if you wanted to.

1309
00:58:38,860 --> 00:58:41,230
But it's not even a
theory in your mind,

1310
00:58:41,230 --> 00:58:44,570
so you won't even encode
the relevant information

1311
00:58:44,570 --> 00:58:47,270
if you don't expect to have
food allergies or gluten

1312
00:58:47,270 --> 00:58:48,940
allergy or the like
if you don't expect

1313
00:58:48,940 --> 00:58:50,023
that to be a likely cause.

1314
00:58:50,023 --> 00:58:52,107
Then you're never going
to actually pay attention,

1315
00:58:52,107 --> 00:58:54,850
and you get infinite amounts of
data on every day and headaches

1316
00:58:54,850 --> 00:58:55,990
and food consumption.

1317
00:58:55,990 --> 00:58:58,390
But you're never going
to notice because you

1318
00:58:58,390 --> 00:59:02,312
don't have a theory in your mind
that that might be the case.

1319
00:59:02,312 --> 00:59:04,770
And so there's a relationship
between attention and memory.

1320
00:59:04,770 --> 00:59:07,768
You only remember or pay
attention to stuff that

1321
00:59:07,768 --> 00:59:09,060
actually are theories of yours.

1322
00:59:09,060 --> 00:59:10,602
And then you'll
remember those things

1323
00:59:10,602 --> 00:59:13,160
that you actually paid
attention to in the first place.

1324
00:59:13,160 --> 00:59:15,355
And so selective
attention, then,

1325
00:59:15,355 --> 00:59:17,230
may have persistent
effects on what we learn.

1326
00:59:17,230 --> 00:59:19,570
Essentially, you could
have unlimited amounts

1327
00:59:19,570 --> 00:59:21,160
of information in front of you.

1328
00:59:21,160 --> 00:59:23,410
And you will never
learn because you never

1329
00:59:23,410 --> 00:59:26,170
just had it on your radar that
this information is actually

1330
00:59:26,170 --> 00:59:27,907
relevant for the issue at hand.

1331
00:59:27,907 --> 00:59:29,740
And Josh Schwartzstein
has a very nice model

1332
00:59:29,740 --> 00:59:31,960
that illustrates that.

1333
00:59:31,960 --> 00:59:35,980
Another example
is from medicine.

1334
00:59:35,980 --> 00:59:38,170
Many women died
from childbed fever

1335
00:59:38,170 --> 00:59:40,360
at hospitals in the
mid-19th century.

1336
00:59:40,360 --> 00:59:42,490
The popular theories
were bad smells

1337
00:59:42,490 --> 00:59:45,540
at the hospital, presence
of male doctors wounded

1338
00:59:45,540 --> 00:59:47,020
the modesty of the mothers.

1339
00:59:47,020 --> 00:59:50,290
That's, of course, nonsense.

1340
00:59:50,290 --> 00:59:51,880
The true explanation is germs.

1341
00:59:51,880 --> 00:59:53,830
Doctors didn't wash their hands.

1342
00:59:53,830 --> 00:59:56,110
And again, it's something--
that's not a tricky thing

1343
00:59:56,110 --> 00:59:58,000
to figure out if you
have the idea, if you

1344
00:59:58,000 --> 01:00:01,377
have the hypothesis or the idea
that this could be important.

1345
01:00:01,377 --> 01:00:03,835
And presumably, some doctors
are washing their hands anyway

1346
01:00:03,835 --> 01:00:06,760
and sometimes are not, and they
would have a lot fewer deaths.

1347
01:00:06,760 --> 01:00:09,040
But nobody was noticing
because, again, just nobody

1348
01:00:09,040 --> 01:00:10,370
was paying attention.

1349
01:00:10,370 --> 01:00:14,050
So once you have the
right theory in mind,

1350
01:00:14,050 --> 01:00:14,930
then you can test it.

1351
01:00:14,930 --> 01:00:17,305
And then, of course, you can
gather the right information

1352
01:00:17,305 --> 01:00:20,890
and pay attention to the
right information and learn.

1353
01:00:20,890 --> 01:00:25,630
Now, the basic model from
Josh Schwartzstein-- beliefs

1354
01:00:25,630 --> 01:00:28,060
today matter for what
is being attended

1355
01:00:28,060 --> 01:00:31,270
to today, which then affect
people's beliefs tomorrow.

1356
01:00:31,270 --> 01:00:35,890
And now if you don't
attend to the right things,

1357
01:00:35,890 --> 01:00:38,470
if your model of the
world is the wrong model,

1358
01:00:38,470 --> 01:00:41,950
you might never learn to
attend to important aspects

1359
01:00:41,950 --> 01:00:43,070
of the world.

1360
01:00:43,070 --> 01:00:44,980
And so then
forecasts and beliefs

1361
01:00:44,980 --> 01:00:48,490
may be biased and persistently
biased in a systematic fashion.

1362
01:00:48,490 --> 01:00:51,430
And you might persistently
misreact and misattribute

1363
01:00:51,430 --> 01:00:55,170
causes to unimportant variables
because you have never

1364
01:00:55,170 --> 01:00:59,927
even considered the idea of
the correct model of the world.

1365
01:00:59,927 --> 01:01:01,510
And so now the paper
I'm going to show

1366
01:01:01,510 --> 01:01:03,500
you is an example of that.

1367
01:01:03,500 --> 01:01:05,750
This is seaweed
farming in Indonesia,

1368
01:01:05,750 --> 01:01:08,050
which is not immediately
obvious why that's related.

1369
01:01:08,050 --> 01:01:09,500
But I'll show you in a second.

1370
01:01:09,500 --> 01:01:12,496
So this is what seaweed
farming looks like.

1371
01:01:12,496 --> 01:01:14,360
I'll give you this
relatively briefly.

1372
01:01:14,360 --> 01:01:16,660
But essentially, it's
like, at some beaches,

1373
01:01:16,660 --> 01:01:19,750
you can grow seaweeds
in these rows.

1374
01:01:19,750 --> 01:01:21,830
And you see these
round things here,

1375
01:01:21,830 --> 01:01:23,350
which are essentially pods.

1376
01:01:23,350 --> 01:01:25,780
And so there are
many factors that are

1377
01:01:25,780 --> 01:01:27,820
important in seaweed farming.

1378
01:01:27,820 --> 01:01:30,340
Essentially, it's like these
pods, which are essentially

1379
01:01:30,340 --> 01:01:31,900
these round things
that you see here,

1380
01:01:31,900 --> 01:01:34,750
they're essentially
grown in different lines.

1381
01:01:34,750 --> 01:01:36,580
And so if you're
seaweed farmers,

1382
01:01:36,580 --> 01:01:39,160
what's really important
is line spacing--

1383
01:01:39,160 --> 01:01:42,520
or potentially important is line
spacing, pod spacing, and pod

1384
01:01:42,520 --> 01:01:44,560
size, how we got these pods.

1385
01:01:44,560 --> 01:01:47,230
Now, how does seaweed
farming actually work?

1386
01:01:47,230 --> 01:01:49,030
I've never done
seaweed farming myself,

1387
01:01:49,030 --> 01:01:50,930
but apparently this
is how it works.

1388
01:01:50,930 --> 01:01:53,470
So farmers use what's called
the bottom method, which

1389
01:01:53,470 --> 01:01:57,490
is they drive wooden stakes
in shallow bottom near shore,

1390
01:01:57,490 --> 01:01:59,560
and they attach lines
to these stakes.

1391
01:01:59,560 --> 01:02:02,290
And they take raw seaweed
from the last harvest

1392
01:02:02,290 --> 01:02:05,350
and cut it into pods, like
these kind of roundish things.

1393
01:02:05,350 --> 01:02:07,480
And the pods are planted
by attaching them

1394
01:02:07,480 --> 01:02:09,850
at a given interval on
the line from the sea.

1395
01:02:09,850 --> 01:02:14,650
And at low tides,
farmers tend the plots.

1396
01:02:14,650 --> 01:02:19,280
And so the seaweed is then
harvested after 30 to 40 days.

1397
01:02:19,280 --> 01:02:22,150
And so now many dimensions
could matter-- the pod size,

1398
01:02:22,150 --> 01:02:24,670
the distance between the lines,
the distance between pods,

1399
01:02:24,670 --> 01:02:25,990
timing, and so on.

1400
01:02:25,990 --> 01:02:28,600
One nice thing about seaweed
is there's many different pods.

1401
01:02:28,600 --> 01:02:30,610
You can actually try
and learn and estimate

1402
01:02:30,610 --> 01:02:33,190
the importance of
these factors over time

1403
01:02:33,190 --> 01:02:35,380
if only you paid attention.

1404
01:02:35,380 --> 01:02:39,910
But the question is, which of
these factors are important?

1405
01:02:39,910 --> 01:02:43,210
And are people paying
attention to those factors?

1406
01:02:46,945 --> 01:02:48,570
And so what the
experiment now is doing

1407
01:02:48,570 --> 01:02:50,638
is there's lots of farmers
in the experiment that

1408
01:02:50,638 --> 01:02:51,555
are quite experienced.

1409
01:02:51,555 --> 01:02:55,320
They have been doing this for a
long time, 18 years of farming.

1410
01:02:55,320 --> 01:02:57,930
Many of them or the vast
majority are literate.

1411
01:02:57,930 --> 01:02:59,820
So these are people
who should really

1412
01:02:59,820 --> 01:03:03,910
know what they're doing in their
seaweed farming experience.

1413
01:03:03,910 --> 01:03:06,840
And so the enumerators
went to visit these farmers

1414
01:03:06,840 --> 01:03:09,480
and to measure and document
their farming methods.

1415
01:03:09,480 --> 01:03:11,960
And when you asked them
about current pod size,

1416
01:03:11,960 --> 01:03:13,950
you just ask them how
big are your pods,

1417
01:03:13,950 --> 01:03:16,932
people just don't
know the answer to.

1418
01:03:16,932 --> 01:03:18,390
They also don't
know the answer to,

1419
01:03:18,390 --> 01:03:19,557
what's the optimum pod size?

1420
01:03:19,557 --> 01:03:21,765
When you just ask them how
big are your pods when you

1421
01:03:21,765 --> 01:03:23,070
plant them, they don't know.

1422
01:03:23,070 --> 01:03:25,028
When you ask them what's
the best way to do it,

1423
01:03:25,028 --> 01:03:25,890
they don't know.

1424
01:03:25,890 --> 01:03:29,012
Importantly, they know exactly
the answer to other questions.

1425
01:03:29,012 --> 01:03:30,720
They know the length
of the typical line.

1426
01:03:30,720 --> 01:03:32,428
They know the distance
between the lines.

1427
01:03:32,428 --> 01:03:34,110
And they also have
very clear ideas

1428
01:03:34,110 --> 01:03:36,210
of what's optimal to do so.

1429
01:03:36,210 --> 01:03:39,300
But they seem to essentially
neglect the pod size dimension

1430
01:03:39,300 --> 01:03:42,770
entirely when they're
thinking about what to do.

1431
01:03:42,770 --> 01:03:45,940
And so in the
experimental trial now,

1432
01:03:45,940 --> 01:03:47,920
they did essentially
different treatments.

1433
01:03:47,920 --> 01:03:49,600
And the treatments
were essentially such

1434
01:03:49,600 --> 01:03:53,350
that farmers were provided
experimental variation

1435
01:03:53,350 --> 01:03:58,030
in the different pod sizes
and essentially induced

1436
01:03:58,030 --> 01:04:00,970
to experiment with pod sizes.

1437
01:04:00,970 --> 01:04:04,645
And in some cases, farmers
were just left on their own.

1438
01:04:04,645 --> 01:04:06,520
They were just induced
to do this experiment.

1439
01:04:06,520 --> 01:04:09,250
And the question was
like, now will farmers

1440
01:04:09,250 --> 01:04:11,540
learn from their own
in this experiment?

1441
01:04:11,540 --> 01:04:14,440
And then in addition, they also
have an information condition

1442
01:04:14,440 --> 01:04:17,620
where they then summarized
that information to the farmers

1443
01:04:17,620 --> 01:04:18,910
in addition.

1444
01:04:18,910 --> 01:04:23,530
So the question is
now, once farmers

1445
01:04:23,530 --> 01:04:25,300
are induced to
experiment, you surely

1446
01:04:25,300 --> 01:04:27,040
would think that
farmers would learn now

1447
01:04:27,040 --> 01:04:29,415
that they're given all the
information that they need on,

1448
01:04:29,415 --> 01:04:31,080
like, here's the
different pod sizes.

1449
01:04:31,080 --> 01:04:32,622
Here's the different
profits that you

1450
01:04:32,622 --> 01:04:36,490
get from the different
types of pods.

1451
01:04:36,490 --> 01:04:39,140
Now they might
learn on their own.

1452
01:04:39,140 --> 01:04:41,770
But if you don't attend to
pod size in the first place,

1453
01:04:41,770 --> 01:04:44,020
if you think it's
irrelevant anyway,

1454
01:04:44,020 --> 01:04:46,390
even if you are induced
to do the experiment,

1455
01:04:46,390 --> 01:04:49,430
you're not going
to learn at all.

1456
01:04:49,430 --> 01:04:52,600
And so then they did essentially
these follow-up surveys.

1457
01:04:52,600 --> 01:04:54,190
And then we'll just skip.

1458
01:04:54,190 --> 01:04:56,830
So now what they do is they
also go through an information

1459
01:04:56,830 --> 01:05:01,358
provision where they essentially
tell farmers the [INAUDIBLE]

1460
01:05:01,358 --> 01:05:02,650
they were doing with everybody.

1461
01:05:02,650 --> 01:05:04,870
But in addition, they
also provided information

1462
01:05:04,870 --> 01:05:07,690
and did some simple
calculations on which

1463
01:05:07,690 --> 01:05:12,300
is the best combination
of pod size and distance.

1464
01:05:12,300 --> 01:05:13,675
Notice that farmers
are literate.

1465
01:05:13,675 --> 01:05:15,410
They're able to do
this on their own.

1466
01:05:15,410 --> 01:05:18,100
But the experiment was just
providing this information

1467
01:05:18,100 --> 01:05:20,590
to farmers explicitly
with a recommendation

1468
01:05:20,590 --> 01:05:24,940
about what's the pod weight and
what is the optimal distance.

1469
01:05:24,940 --> 01:05:26,440
And so now what the
experiment finds

1470
01:05:26,440 --> 01:05:32,140
is large gains from changing
the farming methodology.

1471
01:05:32,140 --> 01:05:34,630
And the trial
participation itself only

1472
01:05:34,630 --> 01:05:36,370
has an insignificant effect.

1473
01:05:36,370 --> 01:05:38,140
That's to say, in
addition, then,

1474
01:05:38,140 --> 01:05:41,470
summarized data for people
has much larger effect.

1475
01:05:41,470 --> 01:05:45,520
Essentially, focusing
people's attention to things

1476
01:05:45,520 --> 01:05:49,480
that they should already know
changes their farming behavior

1477
01:05:49,480 --> 01:05:50,530
a lot.

1478
01:05:50,530 --> 01:05:54,670
And it's essentially
evidence that people were not

1479
01:05:54,670 --> 01:05:56,570
paying attention
in the first place

1480
01:05:56,570 --> 01:05:58,597
even if you make them
participate in a trial

1481
01:05:58,597 --> 01:06:00,430
where they should have
really or could have,

1482
01:06:00,430 --> 01:06:04,940
at least, learned
things on their own.

1483
01:06:04,940 --> 01:06:06,820
There's large
impacts of the trial

1484
01:06:06,820 --> 01:06:09,520
if the data from the trials
is presented to farmers.

1485
01:06:09,520 --> 01:06:11,740
There's no impact of
the trial on dimensions

1486
01:06:11,740 --> 01:06:13,720
that farmers had already
noticed previously,

1487
01:06:13,720 --> 01:06:16,570
presumably because they're
already paying attention

1488
01:06:16,570 --> 01:06:18,642
and were optimizing.

1489
01:06:18,642 --> 01:06:20,100
So now what did we
learn from this?

1490
01:06:20,100 --> 01:06:23,200
And let me quickly
summarize and let you go.

1491
01:06:23,200 --> 01:06:25,110
So we have systematic
learning failures

1492
01:06:25,110 --> 01:06:27,480
even though all the
information was available.

1493
01:06:27,480 --> 01:06:29,430
So farmers simply
did not pay attention

1494
01:06:29,430 --> 01:06:31,380
because they didn't
think that pod size was

1495
01:06:31,380 --> 01:06:34,650
relevant in any
way in their lives.

1496
01:06:34,650 --> 01:06:36,300
And that's a
potential explanation

1497
01:06:36,300 --> 01:06:37,830
why people might
not pay attention

1498
01:06:37,830 --> 01:06:40,560
even to important information
because if your model

1499
01:06:40,560 --> 01:06:43,170
of the world is just
wrong, why would

1500
01:06:43,170 --> 01:06:45,300
you even think that
this is important?

1501
01:06:45,300 --> 01:06:47,880
And the problem here is then
there's never an opportunity

1502
01:06:47,880 --> 01:06:48,690
to actually learn.

1503
01:06:48,690 --> 01:06:50,160
You never even
collect information

1504
01:06:50,160 --> 01:06:51,952
that might reject your
model because you're

1505
01:06:51,952 --> 01:06:55,240
convinced that your model
is right in the first place.

1506
01:06:55,240 --> 01:06:58,290
And so then lack of attention
might generate arbitrarily

1507
01:06:58,290 --> 01:06:59,430
large welfare losses.

1508
01:06:59,430 --> 01:07:02,490
You can really screw up
big time for a long time

1509
01:07:02,490 --> 01:07:06,503
and never really change your
behavior because, essentially,

1510
01:07:06,503 --> 01:07:07,170
you never learn.

1511
01:07:07,170 --> 01:07:09,180
You never even pay
attention to any information

1512
01:07:09,180 --> 01:07:13,400
that you might have available
that says your model is wrong.