1
00:00:00,000 --> 00:00:01,479
[SQUEAKING]

2
00:00:01,479 --> 00:00:03,944
[RUSTLING]

3
00:00:03,944 --> 00:00:10,370
[CLICKING]

4
00:00:10,370 --> 00:00:12,400
FRANK SCHILBACH:
Welcome to Lecture 17.

5
00:00:12,400 --> 00:00:15,050
This lecture talks about
state-dependent preferences,

6
00:00:15,050 --> 00:00:17,836
projection and attribution bias.

7
00:00:21,540 --> 00:00:24,400
The plan for today is to talk
about preference changes.

8
00:00:24,400 --> 00:00:26,940
So we're going to look at
different circumstances

9
00:00:26,940 --> 00:00:30,120
in which people's
preferences, predictably

10
00:00:30,120 --> 00:00:33,090
and sometimes unpredictably,
change over time.

11
00:00:33,090 --> 00:00:36,900
That is to say, sometimes people
are hungry, tired or in pain,

12
00:00:36,900 --> 00:00:39,690
and their preferences
for certain outcomes

13
00:00:39,690 --> 00:00:42,150
may be determined by
that underlying state

14
00:00:42,150 --> 00:00:44,570
of like hunger, for example.

15
00:00:44,570 --> 00:00:47,870
Now, it's pretty clear that
people's preferences change

16
00:00:47,870 --> 00:00:50,930
according to those states, as
in like when somebody is hungry,

17
00:00:50,930 --> 00:00:52,080
they want different things.

18
00:00:52,080 --> 00:00:53,900
They want to eat
different kinds of things

19
00:00:53,900 --> 00:00:57,650
than when they're
not hungry and so on.

20
00:00:57,650 --> 00:01:00,590
But people nonetheless
have trouble predicting

21
00:01:00,590 --> 00:01:02,180
their preference changes.

22
00:01:02,180 --> 00:01:03,950
That is to say, if
somebody is hungry,

23
00:01:03,950 --> 00:01:05,930
they may have a
hard time predicting

24
00:01:05,930 --> 00:01:09,420
how it feels when they're
not hungry and vice versa.

25
00:01:09,420 --> 00:01:12,570
So notice that this lecture
is sort of a continuation

26
00:01:12,570 --> 00:01:16,310
or natural continuation
of our lectures on beliefs

27
00:01:16,310 --> 00:01:17,730
from before.

28
00:01:17,730 --> 00:01:19,910
So before, we looked
at people trying

29
00:01:19,910 --> 00:01:24,650
to learn about unknown
information around them,

30
00:01:24,650 --> 00:01:26,660
and we looked at
various deviations

31
00:01:26,660 --> 00:01:29,140
why people are
not able to learn.

32
00:01:29,140 --> 00:01:31,570
Now we're going to look
at people trying to learn

33
00:01:31,570 --> 00:01:37,440
or failing to learn at times
about their own preferences.

34
00:01:37,440 --> 00:01:39,490
And one particular
case here is what's

35
00:01:39,490 --> 00:01:42,940
called production bias,
which is people's lack of

36
00:01:42,940 --> 00:01:47,750
or inadequate ability to
predict their own preferences,

37
00:01:47,750 --> 00:01:49,360
in particular for
different states

38
00:01:49,360 --> 00:01:50,620
of the world in the future.

39
00:01:50,620 --> 00:01:52,610
I'm going to be more
specific about that,

40
00:01:52,610 --> 00:01:54,610
but mostly, we're going
to talk about projection

41
00:01:54,610 --> 00:01:56,770
bias and the paper by
Loewenstein et al., which

42
00:01:56,770 --> 00:01:57,970
is in the readings.

43
00:01:57,970 --> 00:02:01,510
We're going to talk very
briefly about attribution bias.

44
00:02:01,510 --> 00:02:03,480
That's going to be
covered in recitation.

45
00:02:06,190 --> 00:02:10,539
OK, so although not typically
emphasized in economics,

46
00:02:10,539 --> 00:02:13,540
the simple and obvious fact
is that our preferences

47
00:02:13,540 --> 00:02:15,370
change over time.

48
00:02:15,370 --> 00:02:19,130
There are ways in which
our preferences change.

49
00:02:19,130 --> 00:02:22,630
One is short-term
temporary fluctuations,

50
00:02:22,630 --> 00:02:25,450
and you can call them sort of
state-dependent preferences.

51
00:02:25,450 --> 00:02:26,950
You can call all
of what we're going

52
00:02:26,950 --> 00:02:30,070
to talk about
state-dependent preferences.

53
00:02:30,070 --> 00:02:35,320
Here, the state is a short-run
physiological or psychological

54
00:02:35,320 --> 00:02:39,070
state, for example,
hunger, pain or the like,

55
00:02:39,070 --> 00:02:42,960
or a psychological
state such as mood.

56
00:02:42,960 --> 00:02:45,860
We're going to talk about
all of these in more detail.

57
00:02:45,860 --> 00:02:49,450
Second, there are long-run
systematic changes.

58
00:02:49,450 --> 00:02:52,840
These could be due to own
choices such as addiction.

59
00:02:52,840 --> 00:02:55,030
So like if somebody has
been drinking alcohol

60
00:02:55,030 --> 00:02:57,520
for the last 10
years, their utility

61
00:02:57,520 --> 00:03:00,070
from drinking alcohol
or not drinking alcohol

62
00:03:00,070 --> 00:03:03,320
is quite different from somebody
who has not been drinking ever.

63
00:03:03,320 --> 00:03:07,420
The same is true for
smoking and the like.

64
00:03:07,420 --> 00:03:09,580
Or it could be independent
of one's choices,

65
00:03:09,580 --> 00:03:11,350
for example, aging.

66
00:03:11,350 --> 00:03:13,150
So predictably,
people's preferences

67
00:03:13,150 --> 00:03:15,430
change over time when
they're 20 years old

68
00:03:15,430 --> 00:03:18,320
versus when they're
eight years old.

69
00:03:18,320 --> 00:03:19,990
Notice that for
addiction, there's

70
00:03:19,990 --> 00:03:23,440
also some short-run
temporary fluctuation

71
00:03:23,440 --> 00:03:27,010
such as when people have just
smoked a cigarette versus not.

72
00:03:27,010 --> 00:03:28,990
Their preferences for
an additional cigarette

73
00:03:28,990 --> 00:03:30,800
might be quite different.

74
00:03:30,800 --> 00:03:33,040
But then there's also
long-run systematic changes,

75
00:03:33,040 --> 00:03:34,930
which is like if
somebody has been

76
00:03:34,930 --> 00:03:38,320
smoking a lot during the
last 10 years versus not.

77
00:03:38,320 --> 00:03:40,780
So smoking or
addiction in particular

78
00:03:40,780 --> 00:03:44,170
has both short-term
and long-term

79
00:03:44,170 --> 00:03:50,200
temporary and systematic
changes in people's preferences.

80
00:03:50,200 --> 00:03:54,460
And then there's adaptation to
changes, which happens often

81
00:03:54,460 --> 00:03:57,220
for big changes such
as standard of living

82
00:03:57,220 --> 00:03:59,740
or small changes such
as mug ownership.

83
00:03:59,740 --> 00:04:00,880
What's an example here?

84
00:04:00,880 --> 00:04:04,240
For example, if people win the
lottery, they become happier.

85
00:04:04,240 --> 00:04:07,480
They become actually a lot
happier very quickly, but then

86
00:04:07,480 --> 00:04:11,320
that sort of reverts back to
like their previous standard

87
00:04:11,320 --> 00:04:12,190
of time.

88
00:04:12,190 --> 00:04:14,500
People, a lottery winner
seen like a few days later,

89
00:04:14,500 --> 00:04:17,380
clearly are happier than people
who have not won the lottery,

90
00:04:17,380 --> 00:04:22,840
but their increase in
happiness decreases over time.

91
00:04:22,840 --> 00:04:24,580
So these are three
different types

92
00:04:24,580 --> 00:04:27,240
of sets of preference changes.

93
00:04:27,240 --> 00:04:29,170
You're going to mostly
focus in the lecture

94
00:04:29,170 --> 00:04:32,020
today on, number one,
short-term, temporary

95
00:04:32,020 --> 00:04:33,520
fluctuations.

96
00:04:33,520 --> 00:04:35,810
Now, what am I
talking about here?

97
00:04:35,810 --> 00:04:39,520
So one example here is hunger.

98
00:04:39,520 --> 00:04:43,090
Now, studying hunger is
difficult in experiments

99
00:04:43,090 --> 00:04:46,480
because it's unethical
to starve people.

100
00:04:46,480 --> 00:04:48,430
And you could, of course,
find hungry people

101
00:04:48,430 --> 00:04:49,690
and give them food.

102
00:04:49,690 --> 00:04:52,270
That's in some sense
what was done once

103
00:04:52,270 --> 00:04:55,760
in what's called the Minnesota
starvation experiment.

104
00:04:55,760 --> 00:04:57,970
This is an experiment that
was motivated by the fact

105
00:04:57,970 --> 00:05:01,810
that there was lots of soldiers
in the Second World War

106
00:05:01,810 --> 00:05:05,560
from the US and Europe
and in many other places,

107
00:05:05,560 --> 00:05:08,440
and these soldiers
had been starved

108
00:05:08,440 --> 00:05:10,040
from the war for a long time.

109
00:05:10,040 --> 00:05:12,640
And they came back from the
war to come back to the US,

110
00:05:12,640 --> 00:05:14,360
and one simple
question was, well,

111
00:05:14,360 --> 00:05:17,440
when you have people
who have starved,

112
00:05:17,440 --> 00:05:22,370
have not eaten for a long time
or very little for a long time,

113
00:05:22,370 --> 00:05:26,380
how do you best integrate
them into society,

114
00:05:26,380 --> 00:05:28,620
or how do you best feed them.

115
00:05:28,620 --> 00:05:30,370
Do you feed them very
quickly, like a lot,

116
00:05:30,370 --> 00:05:33,040
or do you sort of gradually
increase their food take over

117
00:05:33,040 --> 00:05:33,980
time?

118
00:05:33,980 --> 00:05:35,860
So then the Army
did some experiments

119
00:05:35,860 --> 00:05:39,190
with healthy volunteers such
as this guy here on the left.

120
00:05:39,190 --> 00:05:43,390
And people were starved,
literally starved

121
00:05:43,390 --> 00:05:47,680
by getting very little food
for a while, such that,

122
00:05:47,680 --> 00:05:50,600
I think after a few weeks,
the guy would look like this.

123
00:05:50,600 --> 00:05:53,830
And then the Army would look
at people's behavior over time.

124
00:05:53,830 --> 00:05:56,450
They were mostly interested
in like once you have somebody

125
00:05:56,450 --> 00:05:57,950
who has not eaten
for quite a while,

126
00:05:57,950 --> 00:05:59,950
once you give them
more food, what's

127
00:05:59,950 --> 00:06:03,100
happening to their
behavior and what's

128
00:06:03,100 --> 00:06:05,440
the best way of doing that.

129
00:06:05,440 --> 00:06:07,300
Coincidentally,
they also reported

130
00:06:07,300 --> 00:06:10,300
what was happening in the
first part of the experiment

131
00:06:10,300 --> 00:06:12,830
when people were
starved to start with.

132
00:06:12,830 --> 00:06:16,557
And so as expected, people
get extremely focused on food

133
00:06:16,557 --> 00:06:17,890
when they don't eat for a while.

134
00:06:17,890 --> 00:06:19,807
They get really, really
interested in learning

135
00:06:19,807 --> 00:06:21,700
about what's going
on with food and when

136
00:06:21,700 --> 00:06:23,480
they get next food and so on.

137
00:06:23,480 --> 00:06:26,620
But as it happens, it seems to
be that their preferences also

138
00:06:26,620 --> 00:06:29,600
change in various other ways.

139
00:06:29,600 --> 00:06:32,050
In particular, people
seem to lose interest

140
00:06:32,050 --> 00:06:34,190
in lots of other activities.

141
00:06:34,190 --> 00:06:36,910
It is reported
decreased alertness,

142
00:06:36,910 --> 00:06:38,980
lack of self-control,
general apathy.

143
00:06:38,980 --> 00:06:41,800
People are just sort of like
almost like a different person

144
00:06:41,800 --> 00:06:43,660
when they are almost
starved compared

145
00:06:43,660 --> 00:06:47,020
to when they have eaten a lot.

146
00:06:47,020 --> 00:06:50,210
Here's some testimonies
from these soldiers.

147
00:06:50,210 --> 00:06:53,050
One of them is, the acquisition
of food-related items

148
00:06:53,050 --> 00:06:54,970
was a reasonable extension
of their heightened

149
00:06:54,970 --> 00:06:55,720
interest in food.

150
00:06:55,720 --> 00:06:59,050
So as we expect, people get
really interested in food.

151
00:06:59,050 --> 00:07:00,910
Much less reasonable
was the buying

152
00:07:00,910 --> 00:07:05,110
of old books, unnecessary
secondhand clothes, knickknacks

153
00:07:05,110 --> 00:07:06,340
and other junk.

154
00:07:06,340 --> 00:07:09,730
Often, after making
such purchases, which

155
00:07:09,730 --> 00:07:12,550
could be afforded
only with sacrifice,

156
00:07:12,550 --> 00:07:16,870
the man would be puzzled as to
why they had bought such more

157
00:07:16,870 --> 00:07:18,640
or less useless articles.

158
00:07:18,640 --> 00:07:20,320
So that's essentially
sort of saying

159
00:07:20,320 --> 00:07:22,660
people got really interested
in certain purchases.

160
00:07:22,660 --> 00:07:24,880
Their preferences towards
those kinds of purchases

161
00:07:24,880 --> 00:07:26,150
seems to have changed.

162
00:07:26,150 --> 00:07:28,150
They seem to have also
some preference reversal.

163
00:07:28,150 --> 00:07:29,830
They seem to want
to buy some things

164
00:07:29,830 --> 00:07:33,250
and then wonder afterwards
why they did so,

165
00:07:33,250 --> 00:07:35,870
which all seems to be the
consequence of being really,

166
00:07:35,870 --> 00:07:38,170
really hungry.

167
00:07:38,170 --> 00:07:41,260
Then came the day when I
lost my will to activity.

168
00:07:41,260 --> 00:07:44,050
I no longer cared to do
anything that required energy,

169
00:07:44,050 --> 00:07:45,700
and days began to drag.

170
00:07:45,700 --> 00:07:48,295
That seems to be in some
ways like a direct effect

171
00:07:48,295 --> 00:07:50,500
of like having a
low caloric intake,

172
00:07:50,500 --> 00:07:53,410
but really, people's
preferences towards physical

173
00:07:53,410 --> 00:07:56,950
or any other activity seems to
have changed quite a bit due

174
00:07:56,950 --> 00:07:58,630
to hunger.

175
00:07:58,630 --> 00:08:00,670
And another one here--

176
00:08:00,670 --> 00:08:04,060
there's nothing that can
hold my interest for long.

177
00:08:04,060 --> 00:08:05,550
I wait for meal times.

178
00:08:05,550 --> 00:08:07,780
And this is where people's
attention seemed to really

179
00:08:07,780 --> 00:08:11,620
be lowered quite a bit, to the
extent that you think attention

180
00:08:11,620 --> 00:08:14,360
affects people's preferences
and decision making

181
00:08:14,360 --> 00:08:15,400
in various ways.

182
00:08:15,400 --> 00:08:19,980
Again, hunger seems to have
this effect on people's choices.

183
00:08:19,980 --> 00:08:22,900
Now, what's a more sort
of like a real world

184
00:08:22,900 --> 00:08:24,850
example for you guys?

185
00:08:24,850 --> 00:08:27,820
Well, that's shopping
on an empty stomach.

186
00:08:27,820 --> 00:08:30,320
There's sort of a classic
study by Nisbett and Kanouse

187
00:08:30,320 --> 00:08:34,159
from 1969, but there's sort
of other evidence later.

188
00:08:34,159 --> 00:08:37,039
And the folk wisdom is that
shopping on an empty stomach

189
00:08:37,039 --> 00:08:42,380
leads people to buy more, and
perhaps also more junk food.

190
00:08:42,380 --> 00:08:45,063
Now, how do you study this?

191
00:08:45,063 --> 00:08:46,730
Well, you can think
about different ways

192
00:08:46,730 --> 00:08:47,990
in which you might do that.

193
00:08:47,990 --> 00:08:50,330
You could sort of
induce people directly

194
00:08:50,330 --> 00:08:54,448
to eat when they're
hungry versus not,

195
00:08:54,448 --> 00:08:56,990
and there's various other ways
in which you could think about

196
00:08:56,990 --> 00:08:58,820
like setting up experiments.

197
00:08:58,820 --> 00:09:01,070
Now, one very
simple way in which

198
00:09:01,070 --> 00:09:05,300
one could study this
issue is by just randomly

199
00:09:05,300 --> 00:09:08,540
giving a sample of individuals
entering the supermarket

200
00:09:08,540 --> 00:09:09,470
a candy bar.

201
00:09:09,470 --> 00:09:10,700
It could be a candy bar.

202
00:09:10,700 --> 00:09:13,380
It could be also a power
bar or just some other food.

203
00:09:13,380 --> 00:09:15,410
And so now, essentially,
what you do then is,

204
00:09:15,410 --> 00:09:17,570
you take a bunch of
people who go shopping.

205
00:09:17,570 --> 00:09:22,010
And some of them might be
hungry and some might not be.

206
00:09:22,010 --> 00:09:24,410
But then the ones that
are hungry in particular,

207
00:09:24,410 --> 00:09:27,650
when they eat like a candy
bar or any other food,

208
00:09:27,650 --> 00:09:29,380
they become less hungry.

209
00:09:29,380 --> 00:09:31,880
And so then we have a treatment
tradition that's less hungry

210
00:09:31,880 --> 00:09:33,630
and a control condition
who is not getting

211
00:09:33,630 --> 00:09:36,500
any food who is more hungry.

212
00:09:36,500 --> 00:09:40,330
And so then that
allows essentially

213
00:09:40,330 --> 00:09:45,550
to look at variation in
whether people are currently

214
00:09:45,550 --> 00:09:47,235
hungry while shopping.

215
00:09:47,235 --> 00:09:48,610
Another thing one
could do is one

216
00:09:48,610 --> 00:09:51,490
could vary the timing of the
last meal before shopping.

217
00:09:51,490 --> 00:09:53,380
There you worry a little
bit, if that's not

218
00:09:53,380 --> 00:09:56,250
experimentally induced,
that the people who

219
00:09:56,250 --> 00:09:59,440
go shopping right after
having a meal versus not

220
00:09:59,440 --> 00:10:02,500
might be quite different
in various other ways.

221
00:10:02,500 --> 00:10:04,000
But you could, in
principle, also

222
00:10:04,000 --> 00:10:06,350
experimentally induce that.

223
00:10:06,350 --> 00:10:09,310
And then you can
monitor how much

224
00:10:09,310 --> 00:10:11,240
and what kind of
food people buy.

225
00:10:11,240 --> 00:10:13,540
So hungry people
tend to buy more.

226
00:10:13,540 --> 00:10:15,440
They just simply buy more food.

227
00:10:15,440 --> 00:10:18,820
They also more tellingly
buy more junk food.

228
00:10:18,820 --> 00:10:21,670
So their preferences really
seem to be different.

229
00:10:21,670 --> 00:10:24,460
To be clear, people tend to buy
more, not just sort of like,

230
00:10:24,460 --> 00:10:28,327
say, I'm going to buy a Power
Bar or, for example, a banana

231
00:10:28,327 --> 00:10:29,410
because I'm really hungry.

232
00:10:29,410 --> 00:10:31,720
I'm going to bite it right
after I leave the store.

233
00:10:31,720 --> 00:10:34,000
But they buy
significantly more, more

234
00:10:34,000 --> 00:10:37,480
than can be explained by just
filling your immediate hunger.

235
00:10:37,480 --> 00:10:40,480
And in particular, they also
seem to change the type of food

236
00:10:40,480 --> 00:10:41,560
that they buy.

237
00:10:41,560 --> 00:10:45,630
In particular, in this
case here, more junk food.

238
00:10:45,630 --> 00:10:47,640
Here is kind of what
this then looks like.

239
00:10:47,640 --> 00:10:50,560
The left guy is full.

240
00:10:50,560 --> 00:10:53,490
He buys lots of lettuce,
diet water, and so on.

241
00:10:53,490 --> 00:10:56,130
The guy on the right is hungry.

242
00:10:56,130 --> 00:11:00,420
He buys all sorts of things,
a lot more as you can see,

243
00:11:00,420 --> 00:11:02,560
and a lot of also
unhealthy food,

244
00:11:02,560 --> 00:11:07,720
including tortilla chips,
cheese, sugar water, and so on.

245
00:11:07,720 --> 00:11:10,485
So next time you go shopping,
you should sort of introspect

246
00:11:10,485 --> 00:11:13,320
and see what your preferences
towards what you buy

247
00:11:13,320 --> 00:11:16,160
have been affected.

248
00:11:16,160 --> 00:11:18,680
Now, another thing you
can think about that

249
00:11:18,680 --> 00:11:21,800
affects your preferences or
underlying state is your sleep.

250
00:11:21,800 --> 00:11:23,390
When you're tired,
the whole world

251
00:11:23,390 --> 00:11:26,520
seems different in various ways.

252
00:11:26,520 --> 00:11:29,540
One example you can think
about is self-control.

253
00:11:29,540 --> 00:11:31,490
And sleep deprivation
is associated

254
00:11:31,490 --> 00:11:33,470
with lack of self-control.

255
00:11:33,470 --> 00:11:35,120
There's sort of
this old literature

256
00:11:35,120 --> 00:11:37,040
that says self-control
is a muscle that

257
00:11:37,040 --> 00:11:38,570
replenishes overnight.

258
00:11:38,570 --> 00:11:41,720
There's some issues
with this literature.

259
00:11:41,720 --> 00:11:45,410
And there's some experiments
that might not necessarily

260
00:11:45,410 --> 00:11:47,870
replicate.

261
00:11:47,870 --> 00:11:51,140
But the idea that
lack of self-control

262
00:11:51,140 --> 00:11:54,680
might be a consequence
of lack of sleep

263
00:11:54,680 --> 00:11:59,602
seems very plausible
in various ways.

264
00:11:59,602 --> 00:12:01,310
Notice that there
could also be causality

265
00:12:01,310 --> 00:12:02,268
in the other direction.

266
00:12:02,268 --> 00:12:05,660
Lack of self-control might
also be a cause of poor sleep.

267
00:12:05,660 --> 00:12:07,520
Now, there's some
evidence that sleep

268
00:12:07,520 --> 00:12:09,800
affects people's preferences.

269
00:12:09,800 --> 00:12:14,450
When people are sleep-deprived,
they tend to gain weight.

270
00:12:14,450 --> 00:12:16,900
Tired people engage more in
what's called cyberloafing.

271
00:12:16,900 --> 00:12:19,130
They surf more on the internet.

272
00:12:19,130 --> 00:12:21,500
There's also some evidence
that lack of sleep

273
00:12:21,500 --> 00:12:23,030
might affect ethical behavior.

274
00:12:23,030 --> 00:12:25,910
People cheat more
when they're tired.

275
00:12:25,910 --> 00:12:29,780
And we have some evidence
in our experiment in India,

276
00:12:29,780 --> 00:12:31,880
which I'm going to show
you in a little bit,

277
00:12:31,880 --> 00:12:34,280
at the end of the semester,
that naps increase savings

278
00:12:34,280 --> 00:12:37,530
and seem to reduce
people's present bias.

279
00:12:37,530 --> 00:12:40,370
So really this is some
evidence of sleep affecting

280
00:12:40,370 --> 00:12:43,440
people's time preferences.

281
00:12:43,440 --> 00:12:47,870
Another example is Badger
et al. and addiction.

282
00:12:47,870 --> 00:12:51,470
This is a very small study
with only 13 subjects.

283
00:12:51,470 --> 00:12:54,170
So you might take all of what's
in there with a little bit

284
00:12:54,170 --> 00:12:59,900
of a grain of salt. But I
think the underlying message

285
00:12:59,900 --> 00:13:03,210
of the findings are correct.

286
00:13:03,210 --> 00:13:04,460
So what did Badger et al. do?

287
00:13:04,460 --> 00:13:06,020
They elicited
people's willingness

288
00:13:06,020 --> 00:13:10,190
to pay for a second dose of
the heroin substitute, BUP.

289
00:13:10,190 --> 00:13:12,530
These are heroin addicts
that are recovering.

290
00:13:12,530 --> 00:13:15,380
And they tend to
have, often, cravings.

291
00:13:15,380 --> 00:13:19,370
And BUP essentially helps
people not to relapse

292
00:13:19,370 --> 00:13:21,020
and not to take heroin again.

293
00:13:21,020 --> 00:13:25,700
And so, at regular intervals,
they get BUP doses.

294
00:13:25,700 --> 00:13:30,860
Now it will be sort of unethical
to withhold people these doses.

295
00:13:30,860 --> 00:13:33,740
But what the
experiment is doing--

296
00:13:33,740 --> 00:13:36,272
and it's still ethically,
perhaps, tricky--

297
00:13:36,272 --> 00:13:38,480
but what they were asking
for is people's willingness

298
00:13:38,480 --> 00:13:43,580
to pay for a second dose of
BUP, the heroin substitute.

299
00:13:43,580 --> 00:13:45,440
So all individuals
in the experiments

300
00:13:45,440 --> 00:13:49,450
regularly receive their
single dose of BUP.

301
00:13:49,450 --> 00:13:54,110
And then there's also an
additional dose they could get.

302
00:13:54,110 --> 00:13:56,632
And then there was some
variation in the experiment.

303
00:13:56,632 --> 00:13:57,590
What was the variation?

304
00:13:57,590 --> 00:14:00,530
There was a variation in
the state of deprivation.

305
00:14:00,530 --> 00:14:03,680
So people were either
asking all the questions

306
00:14:03,680 --> 00:14:07,730
about our willingness to pay
for a second dose of BUP.

307
00:14:07,730 --> 00:14:09,650
Everybody always
got the first dose.

308
00:14:09,650 --> 00:14:12,360
But then there was some
variation in timing.

309
00:14:12,360 --> 00:14:16,852
For some people, the
more deprived condition,

310
00:14:16,852 --> 00:14:18,560
people were asked
about their willingness

311
00:14:18,560 --> 00:14:20,750
to pay for a second
dose two hours

312
00:14:20,750 --> 00:14:23,570
before the scheduled first dose.

313
00:14:23,570 --> 00:14:26,150
And the less deprived
people were asked right

314
00:14:26,150 --> 00:14:28,025
after the scheduled first dose.

315
00:14:28,025 --> 00:14:30,733
Again, everybody's
getting a first dose.

316
00:14:30,733 --> 00:14:32,150
And we're asking
about willingness

317
00:14:32,150 --> 00:14:34,580
to pay for a second dose.

318
00:14:34,580 --> 00:14:36,560
Some people are asked
about this two hours

319
00:14:36,560 --> 00:14:38,753
before the scheduled first
dose, when they're really

320
00:14:38,753 --> 00:14:40,670
sort of deprived and
craving, when they really

321
00:14:40,670 --> 00:14:42,230
want their first dose.

322
00:14:42,230 --> 00:14:45,710
And others are asked when
they're less deprived, right

323
00:14:45,710 --> 00:14:47,960
after the scheduled first dose.

324
00:14:47,960 --> 00:14:50,850
Notice that the second dose
is always held constant.

325
00:14:50,850 --> 00:14:52,910
So everybody always
gets the first dose.

326
00:14:52,910 --> 00:14:55,380
And the second dose is held
constant in the future,

327
00:14:55,380 --> 00:14:57,070
at least in those
kinds of questions.

328
00:14:57,070 --> 00:14:59,360
So really the willingness
to pay should not

329
00:14:59,360 --> 00:15:01,033
depend on your current state.

330
00:15:01,033 --> 00:15:02,450
In the future,
you're going to get

331
00:15:02,450 --> 00:15:04,550
a second dose, which is
once you have already

332
00:15:04,550 --> 00:15:05,870
gotten the first dose.

333
00:15:05,870 --> 00:15:09,830
So whether you ask me like
two hours before or right

334
00:15:09,830 --> 00:15:11,980
after your scheduled dose
should have more affect

335
00:15:11,980 --> 00:15:13,940
on your willingness
to pay in the future

336
00:15:13,940 --> 00:15:16,970
because that experience
is held fixed.

337
00:15:16,970 --> 00:15:19,420
And you're always
getting the first dose.

338
00:15:19,420 --> 00:15:23,900
Now, they are also asked
[INAUDIBLE],, in addition,

339
00:15:23,900 --> 00:15:26,680
the timing of a
potential second dose.

340
00:15:26,680 --> 00:15:29,170
So sometimes the second dose
was later on the same day

341
00:15:29,170 --> 00:15:31,880
and sometimes the second dose
was during the next week.

342
00:15:34,710 --> 00:15:36,590
Now, what do Badger et al. find?

343
00:15:36,590 --> 00:15:38,750
They find that people's
willingness to pay

344
00:15:38,750 --> 00:15:42,020
varies systematically with
the state and the delay.

345
00:15:42,020 --> 00:15:43,040
What's the state here?

346
00:15:43,040 --> 00:15:45,050
The state is the
state of deprivation,

347
00:15:45,050 --> 00:15:47,660
whether people are more
deprived or less deprived.

348
00:15:47,660 --> 00:15:49,100
And then the timing
is essentially

349
00:15:49,100 --> 00:15:53,040
either about today
versus next week.

350
00:15:53,040 --> 00:15:55,730
Now, the median willingness
to pay for the second dose

351
00:15:55,730 --> 00:15:59,090
later today is $50
in the satiated state

352
00:15:59,090 --> 00:16:02,070
and $75 in the deprived state.

353
00:16:02,070 --> 00:16:04,820
And we're going to get back to
this how we think about this,

354
00:16:04,820 --> 00:16:06,140
how we can explain this.

355
00:16:06,140 --> 00:16:09,770
But basically the way we explain
this is that in the deprived

356
00:16:09,770 --> 00:16:12,770
state, people's
willingness to pay

357
00:16:12,770 --> 00:16:16,670
is higher because they really,
really want a first dose.

358
00:16:16,670 --> 00:16:19,370
So people's willingness
to pay for the first dose

359
00:16:19,370 --> 00:16:21,458
is high in the
deprived state and it's

360
00:16:21,458 --> 00:16:23,750
lower in the satiated state
because they already have--

361
00:16:26,420 --> 00:16:29,430
people's willingness to
pay for any additional dose

362
00:16:29,430 --> 00:16:32,090
is high in the deprived state
because they haven't even

363
00:16:32,090 --> 00:16:33,420
gotten the first dose.

364
00:16:33,420 --> 00:16:36,260
And it's lower in
the satiated state.

365
00:16:36,260 --> 00:16:37,940
But what people seem
to be doing now,

366
00:16:37,940 --> 00:16:40,070
they seem to have a
hard time imagining

367
00:16:40,070 --> 00:16:41,900
their utility in the future.

368
00:16:41,900 --> 00:16:45,920
That is to say, in the future,
both people will be satiated.

369
00:16:45,920 --> 00:16:47,835
So what's happening
here is that when

370
00:16:47,835 --> 00:16:49,210
people are in the
deprived state,

371
00:16:49,210 --> 00:16:50,810
they really, really want BUP.

372
00:16:50,810 --> 00:16:54,450
And they're willing
to pay more than when

373
00:16:54,450 --> 00:16:57,530
in the satiated state,
again, for the second dose

374
00:16:57,530 --> 00:16:59,550
in the future.

375
00:16:59,550 --> 00:17:02,040
Now, there's also some
evidence of willingness

376
00:17:02,040 --> 00:17:04,560
to pay for the second
dose next week.

377
00:17:04,560 --> 00:17:06,310
And again, people's
willingness to pay--

378
00:17:06,310 --> 00:17:09,010
and so the previous
one was later today.

379
00:17:09,010 --> 00:17:10,440
Now it's for the next week.

380
00:17:10,440 --> 00:17:14,520
People's willingness to pay
in the deprived state is $60.

381
00:17:14,520 --> 00:17:17,310
In the satiated state, it's $35.

382
00:17:17,310 --> 00:17:20,220
So again, people's
willingness to pay

383
00:17:20,220 --> 00:17:22,440
is higher in the deprived state.

384
00:17:22,440 --> 00:17:24,569
That is to say, when
in the deprived state,

385
00:17:24,569 --> 00:17:27,030
really, really, really
people want BUP.

386
00:17:27,030 --> 00:17:31,170
And they cannot really
imagine that in the next week,

387
00:17:31,170 --> 00:17:36,180
when they're satiated, they
will actually want BUP less.

388
00:17:36,180 --> 00:17:42,930
To be clear, in both the first,
when the dose is later today

389
00:17:42,930 --> 00:17:45,480
and when the dose
is next week, it'll

390
00:17:45,480 --> 00:17:47,520
always be the case that
people have already

391
00:17:47,520 --> 00:17:49,200
received the first dose.

392
00:17:49,200 --> 00:17:51,090
And then the question
is always about what's

393
00:17:51,090 --> 00:17:53,980
your willingness to pay
to receive a second dose.

394
00:17:53,980 --> 00:17:58,090
So when they're in the future,
they will always be satiated.

395
00:17:58,090 --> 00:17:59,910
But people in the
deprived state seem

396
00:17:59,910 --> 00:18:03,450
to be reacting or seem to be
behaving as if, in the future,

397
00:18:03,450 --> 00:18:05,520
they would also be in
the deprived state.

398
00:18:05,520 --> 00:18:07,950
But of course people will
be in the satiated state.

399
00:18:07,950 --> 00:18:11,100
So people seem to overestimate,
when they're in a deprived

400
00:18:11,100 --> 00:18:15,030
state, how much they
would like the second dose

401
00:18:15,030 --> 00:18:18,920
in the satiated
state in the future.

402
00:18:18,920 --> 00:18:21,020
Now, here's another
example by Schelling.

403
00:18:21,020 --> 00:18:24,470
And Schelling is a beautiful
writer in economics.

404
00:18:24,470 --> 00:18:27,410
He's one of the best
writers that economics has.

405
00:18:27,410 --> 00:18:31,490
Granted, there's lots of
bad writing in economics.

406
00:18:31,490 --> 00:18:36,230
But if you want to read
some beautiful writing

407
00:18:36,230 --> 00:18:38,810
by an economist,
Schelling has a bunch

408
00:18:38,810 --> 00:18:42,725
of different books that
are really nice to read

409
00:18:42,725 --> 00:18:45,210
and essays that are
really beautiful.

410
00:18:45,210 --> 00:18:51,950
Now in his paper in
1984, he talks briefly

411
00:18:51,950 --> 00:18:53,690
about a controversial
question, which

412
00:18:53,690 --> 00:18:56,540
is the use of anesthesia
during childbirth.

413
00:18:56,540 --> 00:18:59,270
And why is that controversial?

414
00:18:59,270 --> 00:19:05,150
Well, anesthesia is
reducing the mother's pain

415
00:19:05,150 --> 00:19:07,460
during childbirth.

416
00:19:07,460 --> 00:19:09,860
But there's the potential
for some side effects.

417
00:19:09,860 --> 00:19:13,590
And there's also the potential
that people's experience

418
00:19:13,590 --> 00:19:14,850
is different.

419
00:19:14,850 --> 00:19:17,810
And for both those
reasons, some people

420
00:19:17,810 --> 00:19:21,650
argue that anesthesia
is not warranted

421
00:19:21,650 --> 00:19:25,070
and decide not to do that.

422
00:19:25,070 --> 00:19:28,160
Now, in many cases,
people's preferences

423
00:19:28,160 --> 00:19:30,680
change predictably over time.

424
00:19:30,680 --> 00:19:35,300
First, ex ante, before
the woman gives birth,

425
00:19:35,300 --> 00:19:38,690
many women prefer not
to use anesthesia.

426
00:19:38,690 --> 00:19:41,870
Once they are in
excruciating pain,

427
00:19:41,870 --> 00:19:45,140
they request anesthesia
from their doctor.

428
00:19:45,140 --> 00:19:48,140
And then, if they actually get
the anesthesia from the doctor

429
00:19:48,140 --> 00:19:53,960
ex post, after the child is
born, they regret their choice.

430
00:19:53,960 --> 00:19:56,990
They say, I wish I
hadn't gotten anesthesia.

431
00:19:56,990 --> 00:20:00,410
Now, that pattern
holds even for women

432
00:20:00,410 --> 00:20:03,110
who have given birth before.

433
00:20:03,110 --> 00:20:06,320
And it leads to a bunch
of ethical dilemmas

434
00:20:06,320 --> 00:20:08,210
and legal issues.

435
00:20:08,210 --> 00:20:11,660
For example, to which patient is
a physician obligated, the one

436
00:20:11,660 --> 00:20:16,010
asking for anesthesia or the one
who asked that it be withheld?

437
00:20:16,010 --> 00:20:18,540
The person told the
doctor, I really,

438
00:20:18,540 --> 00:20:20,630
really do not want anesthesia.

439
00:20:20,630 --> 00:20:24,267
Even if I ask you later, please
do not give me anesthesia.

440
00:20:24,267 --> 00:20:26,600
But then, of course, the
person that yells at the doctor

441
00:20:26,600 --> 00:20:28,392
and says, I really,
really want anesthesia.

442
00:20:28,392 --> 00:20:30,080
Give me anesthesia right now.

443
00:20:30,080 --> 00:20:33,170
And so there's essentially
preference reversals.

444
00:20:33,170 --> 00:20:34,670
People seem to want
different things

445
00:20:34,670 --> 00:20:36,530
at different points in time.

446
00:20:36,530 --> 00:20:41,960
And the doctor can only make
one of these two people happy.

447
00:20:41,960 --> 00:20:44,240
And so no matter what
the doctor does, either

448
00:20:44,240 --> 00:20:48,440
if the person gives the person
anesthesia, then person number

449
00:20:48,440 --> 00:20:50,690
2 in my list here above
is going to be happy,

450
00:20:50,690 --> 00:20:53,630
but person number 1 and
person number 3 are unhappy.

451
00:20:53,630 --> 00:20:56,030
Or the person withholds
anesthesia and then person

452
00:20:56,030 --> 00:20:57,050
number 2 is unhappy.

453
00:20:57,050 --> 00:21:01,610
So either way, the doctor cannot
satisfy these different selves

454
00:21:01,610 --> 00:21:02,880
over time.

455
00:21:02,880 --> 00:21:04,790
Now, can the physician
enter a contract

456
00:21:04,790 --> 00:21:07,240
with a patient ex ante, before?

457
00:21:07,240 --> 00:21:09,770
Is it possible to have a
legally binding contract here

458
00:21:09,770 --> 00:21:13,002
when he or she says, I'm not
going to give any anesthesia,

459
00:21:13,002 --> 00:21:15,680
and you told me so?

460
00:21:15,680 --> 00:21:19,580
Do we want policies that
make such contracts possible?

461
00:21:19,580 --> 00:21:21,590
Is that a desirable thing to do?

462
00:21:21,590 --> 00:21:24,350
Is it a legally
required thing to do?

463
00:21:24,350 --> 00:21:27,050
Potentially impossible
to require this legally.

464
00:21:27,050 --> 00:21:28,910
I'm not sure.

465
00:21:28,910 --> 00:21:31,490
But the point here
really is that there

466
00:21:31,490 --> 00:21:35,570
is some misprediction
going on here.

467
00:21:35,570 --> 00:21:39,740
Ex ante, the person says, well,
if I'm in pain, I will be fine.

468
00:21:39,740 --> 00:21:41,450
I'm not going to
need anesthesia.

469
00:21:41,450 --> 00:21:43,825
But then, of course, when the
person is actually in pain,

470
00:21:43,825 --> 00:21:45,170
the person wants anesthesia.

471
00:21:45,170 --> 00:21:49,490
So essentially people seem to
mispredict their preferences

472
00:21:49,490 --> 00:21:50,750
when they're in pain.

473
00:21:50,750 --> 00:21:52,880
Then that leads to all
sorts of trouble here.

474
00:21:52,880 --> 00:21:54,890
I really recommend
to read the paper

475
00:21:54,890 --> 00:21:56,840
by Schelling that's
on the course website

476
00:21:56,840 --> 00:22:01,200
if you want to sort think
about this a little bit more.

477
00:22:01,200 --> 00:22:04,950
Now, so far, we talked
about that preferences

478
00:22:04,950 --> 00:22:09,090
change predictably due to
changes in underlying states.

479
00:22:09,090 --> 00:22:12,450
Now, one of those states, it
would be like pain, hunger,

480
00:22:12,450 --> 00:22:14,260
sleep, and so on.

481
00:22:14,260 --> 00:22:16,770
Now, there's an
additional stylized fact

482
00:22:16,770 --> 00:22:19,500
just that people make systematic
mistakes at predicting

483
00:22:19,500 --> 00:22:20,910
preference changes.

484
00:22:20,910 --> 00:22:25,240
So not only is it that your
preferences are different when

485
00:22:25,240 --> 00:22:28,060
you're tired versus
not, but people are also

486
00:22:28,060 --> 00:22:31,210
systematically mispredicting
that preference change.

487
00:22:31,210 --> 00:22:33,070
Now, what is projection bias?

488
00:22:33,070 --> 00:22:36,760
To be clear, it's the fact that
people underappreciate changes

489
00:22:36,760 --> 00:22:39,940
in their preferences, projecting
their current preferences

490
00:22:39,940 --> 00:22:42,190
onto future preferences.

491
00:22:42,190 --> 00:22:44,260
So projection bias, to
be clear, is not just

492
00:22:44,260 --> 00:22:46,960
some random prediction
but a prediction

493
00:22:46,960 --> 00:22:49,610
with a systematic direction.

494
00:22:49,610 --> 00:22:52,780
So people seem to understand the
direction of preference change

495
00:22:52,780 --> 00:22:54,460
but not the magnitude.

496
00:22:54,460 --> 00:22:56,710
That is to say, when
people are hungry and think

497
00:22:56,710 --> 00:22:59,260
about how it's going to
feel when they're not hungry

498
00:22:59,260 --> 00:23:02,110
anymore, they kind of understand
that eventually they will not

499
00:23:02,110 --> 00:23:04,375
be hungry and that maybe,
suppose you're really hungry

500
00:23:04,375 --> 00:23:05,797
and you really
want potato chips,

501
00:23:05,797 --> 00:23:07,630
you sort of understand
that you don't really

502
00:23:07,630 --> 00:23:10,528
want that many potato
chips when you're full.

503
00:23:10,528 --> 00:23:12,070
Nevertheless, when
people are hungry,

504
00:23:12,070 --> 00:23:15,610
they buy 10 bags of potato
chips as if they were hungry

505
00:23:15,610 --> 00:23:17,530
at least for like
half of their life,

506
00:23:17,530 --> 00:23:20,200
even though they're going to be
hungry for the next hour or two

507
00:23:20,200 --> 00:23:23,840
because then they're going to
go home and go eat and so on.

508
00:23:23,840 --> 00:23:26,920
So people seem to
understand the direction

509
00:23:26,920 --> 00:23:31,010
of people's preferences
but not the magnitude.

510
00:23:31,010 --> 00:23:39,050
Now the underappreciation
of the effects

511
00:23:39,050 --> 00:23:42,110
of hunger on
preferences is perhaps

512
00:23:42,110 --> 00:23:47,460
not the most economically
important part

513
00:23:47,460 --> 00:23:50,810
of projection bias or the
most important application.

514
00:23:50,810 --> 00:23:53,660
But there two reasons to
consider this evidence.

515
00:23:53,660 --> 00:23:55,970
But one, it's perhaps
the clearest evidence

516
00:23:55,970 --> 00:23:57,440
of projection bias.

517
00:23:57,440 --> 00:23:59,690
Second, people have
lots of experience

518
00:23:59,690 --> 00:24:02,640
with changes in their
levels of hunger.

519
00:24:02,640 --> 00:24:05,630
So any misprediction isn't
due to the lack of opportunity

520
00:24:05,630 --> 00:24:08,630
to learn, really, when you
should have probably learned

521
00:24:08,630 --> 00:24:09,800
this over time.

522
00:24:09,800 --> 00:24:11,905
For some other instances--

523
00:24:11,905 --> 00:24:13,280
this is only, for
example, if you

524
00:24:13,280 --> 00:24:16,140
have lots of physical pain, or
for childbirth, for example,

525
00:24:16,140 --> 00:24:21,050
you might say, well, that
seems to be sometimes

526
00:24:21,050 --> 00:24:22,760
even "one in a
lifetime" experience,

527
00:24:22,760 --> 00:24:25,820
where really it's sort of
understandable that people

528
00:24:25,820 --> 00:24:29,550
mispredict their
preferences when

529
00:24:29,550 --> 00:24:32,030
they are in excruciating pain.

530
00:24:32,030 --> 00:24:33,650
But that's understandable
because you

531
00:24:33,650 --> 00:24:35,420
haven't experienced it before.

532
00:24:35,420 --> 00:24:39,800
And even for somebody, a mother
who has given birth to a child

533
00:24:39,800 --> 00:24:45,092
before, that mother might just
misremember to some degree.

534
00:24:45,092 --> 00:24:46,550
But when it comes
to hunger, people

535
00:24:46,550 --> 00:24:49,530
have been hungry so many
times, and full, and so on.

536
00:24:49,530 --> 00:24:52,010
So really you had lots
and lots of opportunities

537
00:24:52,010 --> 00:24:55,370
to learn what your preferences
are when you're hungry

538
00:24:55,370 --> 00:24:56,030
versus not.

539
00:24:56,030 --> 00:24:59,920
So really it shouldn't be about
a lack of opportunity to learn.

540
00:24:59,920 --> 00:25:02,300
Now, people buy more
on an empty stomach.

541
00:25:02,300 --> 00:25:05,660
That can be interpreted as a
manifestation of projection

542
00:25:05,660 --> 00:25:06,260
bias.

543
00:25:06,260 --> 00:25:10,310
Hungry people act as if
their future taste for food

544
00:25:10,310 --> 00:25:13,730
will reflect the current
hunger at least to some degree,

545
00:25:13,730 --> 00:25:16,260
or more so than
it actually does.

546
00:25:16,260 --> 00:25:18,733
But it's not completely clean
evidence of projection bias,

547
00:25:18,733 --> 00:25:20,900
because there could be
always other things going on.

548
00:25:20,900 --> 00:25:23,960
For example, if
you're really hungry,

549
00:25:23,960 --> 00:25:26,195
some things might be
more salient to you.

550
00:25:26,195 --> 00:25:28,220
And there you might
buy certain things

551
00:25:28,220 --> 00:25:31,290
more because you pay more
attention, for example,

552
00:25:31,290 --> 00:25:34,070
to potato chips and really
sort of learn how exciting

553
00:25:34,070 --> 00:25:36,000
potato chips are and so on.

554
00:25:36,000 --> 00:25:39,020
But let's just go
with hunger for now.

555
00:25:39,020 --> 00:25:42,050
There's lots of other evidence
of projection bias in addition

556
00:25:42,050 --> 00:25:44,840
to the hunger example.

557
00:25:44,840 --> 00:25:48,670
Now remember the paper that
we discussed on food choices

558
00:25:48,670 --> 00:25:50,710
by Read and van Leeuwen.

559
00:25:50,710 --> 00:25:52,630
This is the paper
where office workers

560
00:25:52,630 --> 00:25:54,550
are asked to choose
between a healthy snack

561
00:25:54,550 --> 00:25:56,440
and an unhealthy snack.

562
00:25:56,440 --> 00:26:01,300
Now, so far, we had looked at
this as a piece of evidence

563
00:26:01,300 --> 00:26:03,340
for present bias.

564
00:26:03,340 --> 00:26:05,650
We were looking at when
people are choosing

565
00:26:05,650 --> 00:26:07,660
for the present
right now versus when

566
00:26:07,660 --> 00:26:09,910
they choose for the future.

567
00:26:09,910 --> 00:26:14,230
Do people make more
unhealthy choices

568
00:26:14,230 --> 00:26:15,880
when they choose
for the president?

569
00:26:15,880 --> 00:26:17,860
That is to say the
stylized fact that we

570
00:26:17,860 --> 00:26:22,700
found was that when people are
choosing snacks for the future,

571
00:26:22,700 --> 00:26:25,825
they were quite likely
to choose healthy snacks.

572
00:26:25,825 --> 00:26:30,550
You choose salad for the future
as your snack in the afternoon.

573
00:26:30,550 --> 00:26:35,020
And then, you ask again,
right now or today,

574
00:26:35,020 --> 00:26:38,800
what would you like, people
seem to switch their preferences

575
00:26:38,800 --> 00:26:40,480
towards unhealthy snacks.

576
00:26:40,480 --> 00:26:43,480
They'd rather have,
like, chocolates.

577
00:26:43,480 --> 00:26:45,460
And so we interpreted
that, at the time,

578
00:26:45,460 --> 00:26:47,750
as evidence of present bias.

579
00:26:47,750 --> 00:26:50,860
And surely it seems like that
is evidence of present bias.

580
00:26:50,860 --> 00:26:53,830
Now, in addition, now, we're
going to consider variation

581
00:26:53,830 --> 00:26:56,060
in the timing of those choices.

582
00:26:56,060 --> 00:26:58,330
So some people were asked
when they were hungry,

583
00:26:58,330 --> 00:26:59,920
late in the afternoon.

584
00:26:59,920 --> 00:27:03,820
Others were asked when they were
satiated or arguably satiated,

585
00:27:03,820 --> 00:27:06,220
immediately after lunch.

586
00:27:06,220 --> 00:27:08,750
And then the snacks were
to be received in one week.

587
00:27:08,750 --> 00:27:10,690
So that's held
constant, the timing.

588
00:27:10,690 --> 00:27:12,890
Otherwise the timing
is held constant.

589
00:27:12,890 --> 00:27:15,130
So we're always going to
look at choices for snacks

590
00:27:15,130 --> 00:27:16,300
to be received in one week.

591
00:27:19,135 --> 00:27:21,370
In some of the cases,
the snacks were

592
00:27:21,370 --> 00:27:24,850
to be received when people were
hungry or likely hungry, late

593
00:27:24,850 --> 00:27:26,020
in the afternoon.

594
00:27:26,020 --> 00:27:28,250
In some other
cases, the snacks we

595
00:27:28,250 --> 00:27:30,430
received when people
were satiated,

596
00:27:30,430 --> 00:27:32,910
immediately after lunch.

597
00:27:32,910 --> 00:27:35,200
OK, so there's like
four cases here.

598
00:27:35,200 --> 00:27:36,640
There's the timing of choice.

599
00:27:36,640 --> 00:27:40,010
Either people are asked if
they're hungry versus not.

600
00:27:40,010 --> 00:27:41,770
And then the timing
of the receipt

601
00:27:41,770 --> 00:27:45,340
of their snack, which is when
they were hungry versus not.

602
00:27:45,340 --> 00:27:48,250
And the question now is,
are hungry people now

603
00:27:48,250 --> 00:27:51,760
good or bad at predicting
their preferences when they're

604
00:27:51,760 --> 00:27:56,836
satiated, and are satiated
people good or bad

605
00:27:56,836 --> 00:27:59,590
at predicting their preferences
when they're hungry,

606
00:27:59,590 --> 00:28:02,910
in the different state?

607
00:28:02,910 --> 00:28:05,300
So we're going to
interpret here--

608
00:28:05,300 --> 00:28:09,420
so here's a table from Read and
van Leeuwen, table number 1.

609
00:28:09,420 --> 00:28:12,690
It's a bit of a sort
of confusing table

610
00:28:12,690 --> 00:28:14,310
for no good reason.

611
00:28:14,310 --> 00:28:17,560
Let me just walk you very
quickly through that table.

612
00:28:17,560 --> 00:28:20,640
So the rows are
the current hunger

613
00:28:20,640 --> 00:28:23,220
and the columns
are future hunger.

614
00:28:23,220 --> 00:28:25,230
So the first row
is when people were

615
00:28:25,230 --> 00:28:27,220
asked when they were hungry.

616
00:28:27,220 --> 00:28:29,760
So this is late
in the afternoon.

617
00:28:29,760 --> 00:28:32,610
The second row is people were
asked when they were satiated.

618
00:28:32,610 --> 00:28:35,550
That's like right after lunch.

619
00:28:35,550 --> 00:28:37,260
And then the columns are--

620
00:28:37,260 --> 00:28:41,350
again, remember, all of the
choices are for next week.

621
00:28:41,350 --> 00:28:45,570
So the first column is choices
for when people were hungry.

622
00:28:45,570 --> 00:28:47,070
That's late in the afternoon.

623
00:28:47,070 --> 00:28:50,220
And the second column is
when people were satiated.

624
00:28:50,220 --> 00:28:51,880
That's right after lunch.

625
00:28:51,880 --> 00:28:55,380
Remember, again, all
choices are for next week,

626
00:28:55,380 --> 00:28:58,150
so for one week in the future.

627
00:28:58,150 --> 00:29:02,830
Now, we can interpret, now,
the main diagonal of this table

628
00:29:02,830 --> 00:29:05,710
as reflecting people's
true preferences.

629
00:29:05,710 --> 00:29:08,530
That is to say, here's
no projection bias.

630
00:29:08,530 --> 00:29:12,280
People are asked, when hungry,
to predict what they wanted

631
00:29:12,280 --> 00:29:14,860
or to say what they
wanted in the future, when

632
00:29:14,860 --> 00:29:17,980
they were likely going
to be hungry as well.

633
00:29:17,980 --> 00:29:23,350
That is to say, 78% of people
are choosing an unhealthy snack

634
00:29:23,350 --> 00:29:27,070
for the future when
they're hungry-- so

635
00:29:27,070 --> 00:29:30,100
that's in the late afternoon--
for the future case

636
00:29:30,100 --> 00:29:32,680
when they're in
the late afternoon.

637
00:29:32,680 --> 00:29:40,160
In contrast, only 26% of people
choose the unhealthy snack when

638
00:29:40,160 --> 00:29:43,610
they're satiated-- that
is right after lunch--

639
00:29:43,610 --> 00:29:50,250
for when the snack will
be delivered in one week,

640
00:29:50,250 --> 00:29:52,290
right after lunch.

641
00:29:52,290 --> 00:29:54,180
So there's no
projection bias here.

642
00:29:54,180 --> 00:29:56,340
Because essentially this
underlying state, hunger,

643
00:29:56,340 --> 00:29:57,660
is held constant.

644
00:29:57,660 --> 00:30:00,390
Hungry people predicting
for when they're hungry

645
00:30:00,390 --> 00:30:04,740
or asking what they want when
they're hungry, 78% of people

646
00:30:04,740 --> 00:30:06,360
think they want unhealthy stuff.

647
00:30:06,360 --> 00:30:10,830
And again, satiated
people, 26% of people

648
00:30:10,830 --> 00:30:12,690
say they want unhealthy stuff.

649
00:30:12,690 --> 00:30:17,130
So 74% of people say they want
the healthy snack, in contrast,

650
00:30:17,130 --> 00:30:19,960
when they're satiated.

651
00:30:19,960 --> 00:30:21,613
So again, that's
what's written here.

652
00:30:21,613 --> 00:30:23,530
Late in the afternoon,
when people are hungry,

653
00:30:23,530 --> 00:30:26,050
78% of people choose
the unhealthy snack

654
00:30:26,050 --> 00:30:28,300
for the late afternoon
when they will be hungry.

655
00:30:28,300 --> 00:30:31,900
Immediately after lunch, when
they're satiated, 26% of people

656
00:30:31,900 --> 00:30:34,120
choose the unhealthy
snack for immediately

657
00:30:34,120 --> 00:30:38,130
after lunch when they
will be satiated.

658
00:30:38,130 --> 00:30:43,620
Now, then what we see
here is the entries

659
00:30:43,620 --> 00:30:46,230
that are off the main diagonal.

660
00:30:46,230 --> 00:30:49,860
Those data fit the pattern
of projection bias.

661
00:30:49,860 --> 00:30:51,070
Now what do I mean by that?

662
00:30:51,070 --> 00:30:54,090
Let's look first at
people who are hungry

663
00:30:54,090 --> 00:30:56,130
but expect to be satiated.

664
00:30:56,130 --> 00:30:59,080
That is the first row here.

665
00:30:59,080 --> 00:31:00,900
Current hunger is hungry.

666
00:31:00,900 --> 00:31:03,520
That's essentially people are
asked late in the afternoon.

667
00:31:03,520 --> 00:31:06,030
Or think about this as
being likely hungry.

668
00:31:06,030 --> 00:31:10,020
And they're ask about
why what do they

669
00:31:10,020 --> 00:31:16,400
want when they're satiated?

670
00:31:16,400 --> 00:31:19,560
That is right after lunch.

671
00:31:19,560 --> 00:31:24,140
Now, remember, when people
were satiated, 26% of people

672
00:31:24,140 --> 00:31:26,870
said they wanted
an unhealthy snack.

673
00:31:26,870 --> 00:31:29,960
Now, when people are
hungry, 42% of people

674
00:31:29,960 --> 00:31:32,180
say they want an
unhealthy snack.

675
00:31:32,180 --> 00:31:34,650
So that fraction is higher.

676
00:31:34,650 --> 00:31:38,210
So that is very much consistent
with projection bias,

677
00:31:38,210 --> 00:31:44,690
the reason being that, when
people are hungry and say what

678
00:31:44,690 --> 00:31:47,390
they want for themselves
in the hungry state,

679
00:31:47,390 --> 00:31:50,660
well, 78% of people say they
want the unhealthy snack.

680
00:31:50,660 --> 00:31:52,910
If people did not have
any projection bias,

681
00:31:52,910 --> 00:31:56,060
they really should be
saying 26% if you believe

682
00:31:56,060 --> 00:31:59,010
that satiated people
know best what they want

683
00:31:59,010 --> 00:32:00,440
when they will be satiated.

684
00:32:00,440 --> 00:32:03,170
Instead, 42%, a higher
fraction of people,

685
00:32:03,170 --> 00:32:08,350
say they want an unhealthy
snack when predicting

686
00:32:08,350 --> 00:32:12,950
or when answering while
they're hungry for when

687
00:32:12,950 --> 00:32:14,920
they would be satiated.

688
00:32:14,920 --> 00:32:17,710
So people seem to understand the
direction in which their tastes

689
00:32:17,710 --> 00:32:20,680
change as they become satiated,
but they underestimate

690
00:32:20,680 --> 00:32:23,530
the magnitude of this change.

691
00:32:23,530 --> 00:32:27,130
Similarly, people who are
currently satiated but expect

692
00:32:27,130 --> 00:32:31,150
to be hungry underestimate
the effect of hunger

693
00:32:31,150 --> 00:32:34,170
on their preferences
for unhealthy snacks.

694
00:32:34,170 --> 00:32:45,760
Remember, hungry people, when
they choose for next week

695
00:32:45,760 --> 00:32:47,950
when they will be
hungry, 78% of people

696
00:32:47,950 --> 00:32:50,800
say they want the
unhealthy snack.

697
00:32:50,800 --> 00:32:54,340
Now, in contrast, when they're
satiated, 56% of people

698
00:32:54,340 --> 00:32:56,020
choose the unhealthy snack.

699
00:32:56,020 --> 00:32:58,030
So again, that
fraction now deviates

700
00:32:58,030 --> 00:33:01,300
from the fraction when
people are hungry,

701
00:33:01,300 --> 00:33:04,210
which is, again, consistent
with people understand

702
00:33:04,210 --> 00:33:07,240
the direction in which tastes
change as they become hungry,

703
00:33:07,240 --> 00:33:09,920
but they underestimate the
magnitude of this change.

704
00:33:09,920 --> 00:33:12,430
So satiated people think that
when they're in the future,

705
00:33:12,430 --> 00:33:16,750
hungry, they will behave
at least a little bit

706
00:33:16,750 --> 00:33:19,180
like as if they were satiated.

707
00:33:19,180 --> 00:33:21,815
And hungry people seem to
think that, when they're

708
00:33:21,815 --> 00:33:23,440
satiated in the
future, they will still

709
00:33:23,440 --> 00:33:25,450
behave as if they were hungry.

710
00:33:25,450 --> 00:33:29,170
Notice that this
misprediction is partial.

711
00:33:29,170 --> 00:33:44,150
So again, both the 42%
is in between the 26%

712
00:33:44,150 --> 00:33:54,240
and the 42% is in between
the 26% and the 78%.

713
00:33:54,240 --> 00:33:58,800
And again, the 56% is
between the 26% and the 78%.

714
00:33:58,800 --> 00:34:02,490
So there's essentially some
partial misprediction going on

715
00:34:02,490 --> 00:34:04,080
here.

716
00:34:04,080 --> 00:34:08,949
Now, let me show you a number
of other examples of projection

717
00:34:08,949 --> 00:34:09,449
bias.

718
00:34:09,449 --> 00:34:11,616
And then I'm going to write
down a little model that

719
00:34:11,616 --> 00:34:15,060
sort of captures this
information or the information

720
00:34:15,060 --> 00:34:15,929
that's here.

721
00:34:15,929 --> 00:34:19,995
So another example
is catalog orders.

722
00:34:19,995 --> 00:34:21,570
This is Conlin et al.

723
00:34:21,570 --> 00:34:23,850
Imagine there's a cold evening.

724
00:34:23,850 --> 00:34:25,020
It's really cold outside.

725
00:34:25,020 --> 00:34:26,790
Maybe you were just
outside and so on,

726
00:34:26,790 --> 00:34:28,790
and then you are
online shopping.

727
00:34:28,790 --> 00:34:30,540
Notice that in Conlin
et al., really, it's

728
00:34:30,540 --> 00:34:33,239
catalog orders and
not online shopping

729
00:34:33,239 --> 00:34:36,719
because online shopping didn't
quite exist at the time.

730
00:34:36,719 --> 00:34:42,090
But in any case, imagine you
buy, online, a warm jacket.

731
00:34:42,090 --> 00:34:43,830
And again, it's
really cold right now.

732
00:34:43,830 --> 00:34:46,139
You really wish you
had a warm jacket.

733
00:34:46,139 --> 00:34:50,400
But then, when the
warm jacket arrives,

734
00:34:50,400 --> 00:34:53,909
its 30 degrees warmer
when that happens.

735
00:34:53,909 --> 00:34:57,630
Now, that could or could not
be, potentially, a mistake.

736
00:34:57,630 --> 00:35:02,190
In particular, if it's
an unusually cold day,

737
00:35:02,190 --> 00:35:03,852
if people have
projection bias, they

738
00:35:03,852 --> 00:35:06,060
might sort of think that,
well, right now they really

739
00:35:06,060 --> 00:35:07,620
need a warm jacket.

740
00:35:07,620 --> 00:35:09,630
And they can't even
imagine that they might not

741
00:35:09,630 --> 00:35:13,230
need this warm jacket anymore
in a few days from when

742
00:35:13,230 --> 00:35:15,270
it actually arrives
from the order.

743
00:35:15,270 --> 00:35:17,640
And so the hypothesis
is that projection bias

744
00:35:17,640 --> 00:35:21,570
leads to an increase in
purchases of cold-weather items

745
00:35:21,570 --> 00:35:23,460
on cold-weather days.

746
00:35:23,460 --> 00:35:25,980
And not only that,
but controlling

747
00:35:25,980 --> 00:35:28,470
for receive-day
temperature, the likelihood

748
00:35:28,470 --> 00:35:30,540
of returning a
cold-weather object

749
00:35:30,540 --> 00:35:33,660
is higher when the day
of ordering was cold.

750
00:35:33,660 --> 00:35:37,770
So not only people are buying
more cold-weather items

751
00:35:37,770 --> 00:35:39,105
on cold-weather days--

752
00:35:39,105 --> 00:35:40,230
that seems very reasonable.

753
00:35:40,230 --> 00:35:42,942
Because you might just say, I
always wanted a jacket anyway.

754
00:35:42,942 --> 00:35:43,650
It's really cold.

755
00:35:43,650 --> 00:35:46,620
And that just reminds me
that I really need one.

756
00:35:46,620 --> 00:35:49,463
Or, for example, it's really
rainy and you buy an umbrella.

757
00:35:49,463 --> 00:35:50,880
Because again, you
just remembered

758
00:35:50,880 --> 00:35:52,880
that you really needed
one or that you forgot it

759
00:35:52,880 --> 00:35:54,900
somewhere and so on.

760
00:35:54,900 --> 00:35:56,790
The telling part here
is that the likelihood

761
00:35:56,790 --> 00:36:00,120
of returning the
cold-weather object

762
00:36:00,120 --> 00:36:04,410
is higher when the day
of ordering was cold.

763
00:36:04,410 --> 00:36:07,050
And that's very much consistent
with projection bias.

764
00:36:07,050 --> 00:36:10,260
People, on
cold-weather days, act

765
00:36:10,260 --> 00:36:14,100
as if the future were only
full of cold-weather days.

766
00:36:14,100 --> 00:36:19,410
And therefore their
preferences would be such

767
00:36:19,410 --> 00:36:22,710
that they would always
want to put weather items.

768
00:36:22,710 --> 00:36:28,170
Or put differently, people
act as if their preferences

769
00:36:28,170 --> 00:36:31,170
for cold-weather items-- they
really want a warm jacket--

770
00:36:31,170 --> 00:36:33,330
would not change even
if, in the future,

771
00:36:33,330 --> 00:36:34,920
there will be warm days.

772
00:36:34,920 --> 00:36:37,440
Notice that, in some of
this evidence, what's

773
00:36:37,440 --> 00:36:42,840
hard to distinguish is
predicting people's preferences

774
00:36:42,840 --> 00:36:46,900
change in the future versus
predicting future states.

775
00:36:46,900 --> 00:36:50,580
So if somebody predicted that
it's going to be cold forever,

776
00:36:50,580 --> 00:36:52,500
who would act in
the exact same way

777
00:36:52,500 --> 00:36:56,700
as if you predicted that your
preferences are such that you

778
00:36:56,700 --> 00:36:59,640
always want warm jackets,
you're always going to be cold.

779
00:36:59,640 --> 00:37:01,710
Now, projection bias
is about the latter.

780
00:37:01,710 --> 00:37:03,450
It's about people's preferences.

781
00:37:03,450 --> 00:37:07,410
But often it's actually
hard to separate predictions

782
00:37:07,410 --> 00:37:09,300
of probabilities
of how likely it

783
00:37:09,300 --> 00:37:11,340
is that it's going to
be cold in the future

784
00:37:11,340 --> 00:37:15,970
from predictions of
people's utility.

785
00:37:15,970 --> 00:37:21,850
Another example is car purchases
on Sunday versus radio or days

786
00:37:21,850 --> 00:37:23,710
imagine it's a nice,
sunny day and you

787
00:37:23,710 --> 00:37:27,910
go car shopping as you do it.

788
00:37:27,910 --> 00:37:33,020
And the car dealer
offers you an Audi TT.

789
00:37:33,020 --> 00:37:35,740
You know, "Why, I guess I
never thought of the Audi TT."

790
00:37:35,740 --> 00:37:37,120
You go for a test drive.

791
00:37:37,120 --> 00:37:39,043
And the wind rips
through your hair.

792
00:37:39,043 --> 00:37:40,210
It's really sunny and great.

793
00:37:40,210 --> 00:37:45,290
And having this car during a
sunny day is really amazing.

794
00:37:45,290 --> 00:37:48,370
So you would love to
buy a car like this.

795
00:37:48,370 --> 00:37:50,800
Why not?

796
00:37:50,800 --> 00:37:54,670
Imagine, in contrast, car
purchases on an icy day.

797
00:37:54,670 --> 00:37:56,950
Imagine you're car shopping
around the time of a freak

798
00:37:56,950 --> 00:37:57,555
snowstorm.

799
00:37:57,555 --> 00:37:58,930
You kind of wanted
to have a car,

800
00:37:58,930 --> 00:38:02,470
and somehow it's a snowstorm
when you go car shopping.

801
00:38:02,470 --> 00:38:07,420
Now, the car dealer says, how
about this grand Jeep Grand

802
00:38:07,420 --> 00:38:07,930
Cherokee.

803
00:38:07,930 --> 00:38:11,050
And you might say, well,
I guess I never thought

804
00:38:11,050 --> 00:38:12,790
of the Jeep Grand Cherokee.

805
00:38:12,790 --> 00:38:14,620
Why not try it?

806
00:38:14,620 --> 00:38:17,677
You go for a test drive and you
gain traction on the black ice

807
00:38:17,677 --> 00:38:19,010
and you jump the curb with ease.

808
00:38:19,010 --> 00:38:20,830
It's just amazing.

809
00:38:20,830 --> 00:38:25,790
Now, you might just say, well,
I would love a car like this.

810
00:38:25,790 --> 00:38:28,630
Now, what people
seem to be forgetting

811
00:38:28,630 --> 00:38:30,970
in both of these examples
is that it might just

812
00:38:30,970 --> 00:38:36,760
happen to be a really bad snow
day on that particular day.

813
00:38:36,760 --> 00:38:38,830
Or it might have just
been really sunny

814
00:38:38,830 --> 00:38:40,180
on that particular day.

815
00:38:40,180 --> 00:38:42,310
And things like
change over time.

816
00:38:42,310 --> 00:38:45,340
And people might not understand
that their preferences

817
00:38:45,340 --> 00:38:49,960
for the Audi versus
the Jeep might change

818
00:38:49,960 --> 00:38:52,700
with the weather over time.

819
00:38:52,700 --> 00:38:56,320
And that's exactly what
Busse et al. are looking at.

820
00:38:56,320 --> 00:38:59,770
They're looking at how does
weather impact automobile

821
00:38:59,770 --> 00:39:01,700
purchases.

822
00:39:01,700 --> 00:39:09,070
And in particular the
classical prediction

823
00:39:09,070 --> 00:39:11,900
is that the weather, on
the day of a car purchase,

824
00:39:11,900 --> 00:39:15,280
should have no influence
on the type of car bought.

825
00:39:15,280 --> 00:39:18,790
So now what Busse et al. look
at is do idiosyncratic weather

826
00:39:18,790 --> 00:39:22,240
conditions, controlling for time
of year and predict car sales

827
00:39:22,240 --> 00:39:25,870
and then, in particular,
also returns of those cars.

828
00:39:25,870 --> 00:39:28,000
And they look at
two types of cars,

829
00:39:28,000 --> 00:39:29,887
in particular
convertibles and then

830
00:39:29,887 --> 00:39:31,720
they look at four-wheel
drives like the Jeep

831
00:39:31,720 --> 00:39:33,010
that I just showed you.

832
00:39:33,010 --> 00:39:36,700
And the key part here is to
look at idiosyncratic weather

833
00:39:36,700 --> 00:39:37,270
conditions.

834
00:39:37,270 --> 00:39:40,150
That is to say, controlling
for time of the year.

835
00:39:40,150 --> 00:39:44,890
Suppose it's an unusually warm
and unusually cold or icy day.

836
00:39:44,890 --> 00:39:48,850
Does that predict people's
car sales or purchases

837
00:39:48,850 --> 00:39:52,460
and particularly also the
returns of those cars?

838
00:39:52,460 --> 00:39:56,210
Now, the buying patterns
are very much consistent

839
00:39:56,210 --> 00:39:57,500
with projection bias.

840
00:39:57,500 --> 00:40:00,380
People buy more convertibles
on good-weather days

841
00:40:00,380 --> 00:40:04,710
and more four-wheel drives
on bad-weather days.

842
00:40:04,710 --> 00:40:08,270
And again, that
evidence by itself

843
00:40:08,270 --> 00:40:09,740
is not necessarily
projection bias.

844
00:40:09,740 --> 00:40:12,230
It could just be a good-weather
day and you kind of always

845
00:40:12,230 --> 00:40:14,300
wanted to buy a really
nice car, and therefore

846
00:40:14,300 --> 00:40:16,220
you just do it on
a good-weather day.

847
00:40:16,220 --> 00:40:20,600
But people are also more likely
to return their convertible

848
00:40:20,600 --> 00:40:23,120
if they bought it in
the good-weather day.

849
00:40:23,120 --> 00:40:26,000
And they're likely to return
their four-wheel drive if they

850
00:40:26,000 --> 00:40:27,620
bought it on a bad-weather day.

851
00:40:27,620 --> 00:40:29,690
And that's very much
consistent with evidence

852
00:40:29,690 --> 00:40:30,980
of projection bias.

853
00:40:30,980 --> 00:40:33,920
On a good-weather day,
you think your convertible

854
00:40:33,920 --> 00:40:35,600
is going to be amazing forever.

855
00:40:35,600 --> 00:40:36,808
You're going to love driving.

856
00:40:36,808 --> 00:40:38,350
And even if it's
going to be raining,

857
00:40:38,350 --> 00:40:39,440
it's going to be amazing.

858
00:40:39,440 --> 00:40:41,440
But of course, if it's
raining, that convertible

859
00:40:41,440 --> 00:40:44,337
is really not a car
that you want to have.

860
00:40:44,337 --> 00:40:46,670
Or is really cold outside,
that's really not a great car

861
00:40:46,670 --> 00:40:47,510
to have.

862
00:40:47,510 --> 00:40:50,750
Similarly, if you have
a four-wheel drive

863
00:40:50,750 --> 00:40:52,790
and it's really warm
and nice outside,

864
00:40:52,790 --> 00:40:56,360
you'd probably much rather have
a lighter car or potentially

865
00:40:56,360 --> 00:40:58,220
even a convertible.

866
00:40:58,220 --> 00:41:00,980
Now, one important
question here is now,

867
00:41:00,980 --> 00:41:03,620
in any of the behavioral
biases that we

868
00:41:03,620 --> 00:41:07,410
discussed is how does the
market react to such behavior.

869
00:41:07,410 --> 00:41:10,340
So one tricky part here is that
you want to be kind of careful.

870
00:41:10,340 --> 00:41:11,840
So on one hand, you
might say, well,

871
00:41:11,840 --> 00:41:14,128
let's exploit this
kind of behavior.

872
00:41:14,128 --> 00:41:15,170
And I think that's right.

873
00:41:15,170 --> 00:41:17,390
You might get people to
purchase stuff that they

874
00:41:17,390 --> 00:41:19,190
don't necessarily want.

875
00:41:19,190 --> 00:41:21,710
And surely that's
happening in some cases.

876
00:41:21,710 --> 00:41:25,220
But to the extent that people
can just return these items,

877
00:41:25,220 --> 00:41:27,225
that's actually not
necessarily a great idea.

878
00:41:27,225 --> 00:41:30,392
And particularly if
you're a car salesperson

879
00:41:30,392 --> 00:41:32,600
and you sell people stuff
and then you just come back

880
00:41:32,600 --> 00:41:34,520
and you have to deal with
them, and they're unhappy,

881
00:41:34,520 --> 00:41:35,820
and maybe, then, at
the end of the day,

882
00:41:35,820 --> 00:41:38,112
you just return the car and
actually don't buy anything

883
00:41:38,112 --> 00:41:40,430
eventually because
they're unhappy,

884
00:41:40,430 --> 00:41:42,190
that's really not great for you.

885
00:41:42,190 --> 00:41:45,350
Plus it wastes a lot
of time and effort.

886
00:41:45,350 --> 00:41:48,440
On the other hand,
sometimes people

887
00:41:48,440 --> 00:41:49,880
might just not return things.

888
00:41:49,880 --> 00:41:51,860
And they might just get
stuck with their car

889
00:41:51,860 --> 00:41:53,640
and just leave it at that.

890
00:41:53,640 --> 00:41:56,720
And then it might be a good
idea to exploit this behavior,

891
00:41:56,720 --> 00:41:59,640
particularly on a
good-weather day,

892
00:41:59,640 --> 00:42:02,810
you might want to try to
sell people convertibles.

893
00:42:02,810 --> 00:42:05,540
Another thing you could
potentially exploit

894
00:42:05,540 --> 00:42:08,790
is essentially people just
have higher willingness to pay.

895
00:42:08,790 --> 00:42:10,460
Suppose somebody
really, really wants

896
00:42:10,460 --> 00:42:13,430
to buy a convertible anyway,
regardless of the weather.

897
00:42:13,430 --> 00:42:16,220
But on a good-weather day,
their willingness to pay

898
00:42:16,220 --> 00:42:17,297
might be higher.

899
00:42:17,297 --> 00:42:19,880
And so you might be able to sell
them additional stuff for you

900
00:42:19,880 --> 00:42:23,338
might be able to sell them maybe
a nicer version of the car,

901
00:42:23,338 --> 00:42:25,130
which probably they're
not going to return.

902
00:42:25,130 --> 00:42:27,630
Because once they have it,
they're just going to keep it.

903
00:42:27,630 --> 00:42:35,000
And so you might want to exploit
that if you are a company.

904
00:42:37,700 --> 00:42:46,680
There might also be some agency
issues within the sales shop,

905
00:42:46,680 --> 00:42:52,080
which is like if the salesperson
really wants to sell things

906
00:42:52,080 --> 00:42:54,390
and is incentivized
to sell things,

907
00:42:54,390 --> 00:42:57,300
and that's regardless
of the returns,

908
00:42:57,300 --> 00:42:59,080
then you might get
into situations where,

909
00:42:59,080 --> 00:43:01,650
on sunny days, lots of
convertibles are being sold

910
00:43:01,650 --> 00:43:03,190
but they're all being returned.

911
00:43:03,190 --> 00:43:04,860
So if you are a
company, you kind of

912
00:43:04,860 --> 00:43:08,580
want to set the incentives right
that the car salesman gets only

913
00:43:08,580 --> 00:43:12,210
rewarded for stuff that's not
returned because it precisely

914
00:43:12,210 --> 00:43:15,660
gets you things that will not
make you happy or profitable

915
00:43:15,660 --> 00:43:17,230
in the long run.

916
00:43:17,230 --> 00:43:20,070
Now, there's another--
so as a customer,

917
00:43:20,070 --> 00:43:22,920
now, how might you take
advantage of production bias?

918
00:43:22,920 --> 00:43:27,210
Well, there's an
article here that you

919
00:43:27,210 --> 00:43:29,700
can look at that argues
that winter is the best

920
00:43:29,700 --> 00:43:31,960
time to buy a convertible.

921
00:43:31,960 --> 00:43:34,930
And so the reason is,
of course, convertibles

922
00:43:34,930 --> 00:43:38,530
are often cheaper
in winter in part

923
00:43:38,530 --> 00:43:42,935
because the car dealer might
have storage costs, the like.

924
00:43:42,935 --> 00:43:44,560
So the convertible
is actually cheaper.

925
00:43:44,560 --> 00:43:46,690
Or just because people
are taking advantage

926
00:43:46,690 --> 00:43:49,630
of in the summer when they
really want convertibles.

927
00:43:49,630 --> 00:43:51,673
Now, you might say,
well, let's just buy

928
00:43:51,673 --> 00:43:52,840
a convertible in the winter.

929
00:43:52,840 --> 00:43:54,730
And I think that's,
in principle, right

930
00:43:54,730 --> 00:43:56,860
if you have the time to
wait until the summer

931
00:43:56,860 --> 00:43:58,733
or have a long time horizon.

932
00:43:58,733 --> 00:44:00,400
Now, you want to be
a little bit careful

933
00:44:00,400 --> 00:44:03,670
with exploiting these
market conditions

934
00:44:03,670 --> 00:44:08,200
as a classical agent,
which is manifested

935
00:44:08,200 --> 00:44:12,760
by some of my
colleagues who shall

936
00:44:12,760 --> 00:44:15,820
be unnamed who wanted to
buy a convertible sometime

937
00:44:15,820 --> 00:44:16,690
in the summer.

938
00:44:16,690 --> 00:44:18,670
They got really excited
about the convertible.

939
00:44:18,670 --> 00:44:21,250
But then they realized,
well, winter is actually

940
00:44:21,250 --> 00:44:23,200
the time to buy a
convertible because it's

941
00:44:23,200 --> 00:44:24,317
going to be cheaper.

942
00:44:24,317 --> 00:44:25,900
So then they decided,
well, let's just

943
00:44:25,900 --> 00:44:29,995
wait until the winter and
then buy the convertible then.

944
00:44:29,995 --> 00:44:31,870
Of course, what then
unfortunately happened--

945
00:44:31,870 --> 00:44:32,745
or maybe fortunately.

946
00:44:32,745 --> 00:44:34,420
It depends on the perspective.

947
00:44:34,420 --> 00:44:36,250
Once winter arrived,
they actually

948
00:44:36,250 --> 00:44:38,830
didn't want a
convertible anymore.

949
00:44:38,830 --> 00:44:41,950
Because a convertible in
the winter is really not

950
00:44:41,950 --> 00:44:42,700
a lot of fun.

951
00:44:42,700 --> 00:44:45,610
So projection bias
perhaps also kicked in

952
00:44:45,610 --> 00:44:47,680
with my colleagues,
where they then perhaps

953
00:44:47,680 --> 00:44:49,480
should have just
bought the convertible

954
00:44:49,480 --> 00:44:51,880
and then predicted that, in the
summer, they really wanted it.

955
00:44:51,880 --> 00:44:53,047
But instead they did buy it.

956
00:44:53,047 --> 00:44:56,750
Because wants a
convertible in the winter?

957
00:44:56,750 --> 00:44:59,320
So when exploiting behavioral
biases from others,

958
00:44:59,320 --> 00:45:06,760
you want to be careful that you
may be affected by it yourself.

959
00:45:06,760 --> 00:45:08,620
Let me give you one
last example, which

960
00:45:08,620 --> 00:45:12,520
is this example by Van
Boven and Loewenstein,

961
00:45:12,520 --> 00:45:14,710
which is about thirst.

962
00:45:14,710 --> 00:45:18,458
Here, visitors are asked
before or after vigorous

963
00:45:18,458 --> 00:45:20,500
cardiovascular workouts--
so people who are going

964
00:45:20,500 --> 00:45:21,430
to the gym--

965
00:45:21,430 --> 00:45:23,410
to complete a short survey.

966
00:45:23,410 --> 00:45:24,950
And the survey was as follows.

967
00:45:24,950 --> 00:45:29,200
Imagine that three vacationers
in Colorado this past August

968
00:45:29,200 --> 00:45:31,750
embarked on a short 6-mile hike.

969
00:45:31,750 --> 00:45:33,580
As the day wore
on, they realized

970
00:45:33,580 --> 00:45:36,490
that they were hopelessly lost.

971
00:45:36,490 --> 00:45:40,000
Worse, because they had packed
lightly for a short hike,

972
00:45:40,000 --> 00:45:43,640
they had not carried much
in the way of food or water.

973
00:45:43,640 --> 00:45:45,640
And what people were
asked, then, about-- and so

974
00:45:45,640 --> 00:45:49,640
they were essentially
given this story.

975
00:45:49,640 --> 00:45:56,200
And then they were asked
to do the following.

976
00:45:56,200 --> 00:45:58,840
In the space below, please
take the perspective of one

977
00:45:58,840 --> 00:46:01,480
of the three hikers and
describe your situation--

978
00:46:01,480 --> 00:46:03,340
how you got into it,
how you feel now,

979
00:46:03,340 --> 00:46:07,450
both physically and mentally,
and what you are hoping

980
00:46:07,450 --> 00:46:08,740
will happen."

981
00:46:08,740 --> 00:46:12,580
And now what happens and what
the evidence of Van Boven

982
00:46:12,580 --> 00:46:16,060
and Loewenstein says is that
thirsty subjects have way more

983
00:46:16,060 --> 00:46:18,650
empathy for others' thirst.

984
00:46:18,650 --> 00:46:20,890
So when you look at
the different outcomes,

985
00:46:20,890 --> 00:46:23,170
before and after exercising--

986
00:46:23,170 --> 00:46:27,250
remember, after exercising
is when people are thirsty--

987
00:46:27,250 --> 00:46:32,740
thirst was mentioned
before hunger in the assay,

988
00:46:32,740 --> 00:46:34,960
for thirsty people much more.

989
00:46:34,960 --> 00:46:36,160
Thirst was unpleasant.

990
00:46:36,160 --> 00:46:38,260
For hikers, it was
mentioned much more.

991
00:46:38,260 --> 00:46:40,780
Hikers would regret
not packing water.

992
00:46:40,780 --> 00:46:42,970
Thirst more unpleasant
for the self.

993
00:46:42,970 --> 00:46:46,492
Oneself would regret
more not packing water.

994
00:46:46,492 --> 00:46:47,950
So all of these
items that are sort

995
00:46:47,950 --> 00:46:50,770
of like thirst-related,
people, when

996
00:46:50,770 --> 00:46:53,020
they were thirsty
after exercising

997
00:46:53,020 --> 00:46:57,760
were way more
focused on than not.

998
00:46:57,760 --> 00:47:00,730
So it is as if people have
projection bias in the sense

999
00:47:00,730 --> 00:47:05,170
people understand much more
that the condition of thirst

1000
00:47:05,170 --> 00:47:08,080
is really bad when
they're thirsty

1001
00:47:08,080 --> 00:47:11,680
compared to when they are not.

1002
00:47:11,680 --> 00:47:14,170
OK, so let me take stock
of what we discussed.

1003
00:47:14,170 --> 00:47:16,990
So I showed you some
evidence of projection bias

1004
00:47:16,990 --> 00:47:21,100
for many short-run
changes in preferences,

1005
00:47:21,100 --> 00:47:23,110
and at least for some of them.

1006
00:47:23,110 --> 00:47:26,290
There's hunger, thirst, pain,
sleep, weather, addiction.

1007
00:47:26,290 --> 00:47:27,670
There's other evidence as well--

1008
00:47:27,670 --> 00:47:30,543
arousal, anger, sadness,
and so on-- when

1009
00:47:30,543 --> 00:47:31,960
people aren't
certain states, they

1010
00:47:31,960 --> 00:47:34,900
have trouble to predict their
preferences when they're not

1011
00:47:34,900 --> 00:47:36,820
in that state.

1012
00:47:36,820 --> 00:47:39,280
Now, one key
question that I don't

1013
00:47:39,280 --> 00:47:42,790
have a great answer for you
is why do people not learn?

1014
00:47:42,790 --> 00:47:45,390
People have had
lots of experience

1015
00:47:45,390 --> 00:47:47,140
with a lot of these
changes, in particular

1016
00:47:47,140 --> 00:47:50,420
when you think about sleep
or hunger or the like.

1017
00:47:50,420 --> 00:47:52,150
But even for addiction,
when people just

1018
00:47:52,150 --> 00:47:55,460
have smoked a cigarette versus
not, that happens all the time.

1019
00:47:55,460 --> 00:47:57,460
So they really should
have learned, over time,

1020
00:47:57,460 --> 00:48:00,700
to predict their preference.

1021
00:48:00,700 --> 00:48:03,330
So the misprediction is
really not due to a lack

1022
00:48:03,330 --> 00:48:04,930
of opportunity to learn.

1023
00:48:04,930 --> 00:48:06,930
Yet people really seem
to believe, all the time,

1024
00:48:06,930 --> 00:48:10,450
that this time is different,
over and over again.

1025
00:48:10,450 --> 00:48:12,810
So I think that's just
a very deep cognitive

1026
00:48:12,810 --> 00:48:14,430
bias in some ways
in the sense of,

1027
00:48:14,430 --> 00:48:16,830
when people are affected
by certain visceral

1028
00:48:16,830 --> 00:48:19,710
or other aspects that
affect their preferences,

1029
00:48:19,710 --> 00:48:23,220
it's really hard to
imagine how they might feel

1030
00:48:23,220 --> 00:48:27,767
when that visceral influence
is not at play anymore

1031
00:48:27,767 --> 00:48:28,350
in the future.

1032
00:48:31,920 --> 00:48:35,720
In addition to people
underestimating

1033
00:48:35,720 --> 00:48:37,790
short-run changes
in preferences,

1034
00:48:37,790 --> 00:48:41,270
people also underestimate
adaptation to long-run changes.

1035
00:48:41,270 --> 00:48:42,980
So Dan Gilbert, if
you're interested,

1036
00:48:42,980 --> 00:48:46,610
has very nice work,
including a book

1037
00:48:46,610 --> 00:48:48,410
called Stumbling on Happiness.

1038
00:48:48,410 --> 00:48:51,290
And he gives many examples
of the underestimation

1039
00:48:51,290 --> 00:48:55,530
of adaptation, which
they call immune neglect.

1040
00:48:55,530 --> 00:49:02,760
And so one example is how does
a positive or negative tenure

1041
00:49:02,760 --> 00:49:05,130
decision affect well-being?

1042
00:49:05,130 --> 00:49:09,300
And so they ask current
assistant professors

1043
00:49:09,300 --> 00:49:11,490
at the University of
Texas to forecast.

1044
00:49:11,490 --> 00:49:14,400
And then they ask, as
well, former University

1045
00:49:14,400 --> 00:49:17,070
of Texas assistant
professors to recall.

1046
00:49:17,070 --> 00:49:20,430
And as a current assistant
professor like myself,

1047
00:49:20,430 --> 00:49:23,700
if you forecast, if there's
a negative tenure decision,

1048
00:49:23,700 --> 00:49:24,810
life will be terrible.

1049
00:49:24,810 --> 00:49:29,880
And how are we ever going to
live after being denied tenure?

1050
00:49:32,400 --> 00:49:37,830
And so when you look at
what's actually happening,

1051
00:49:37,830 --> 00:49:41,850
professors seem to be relatively
accurate in predicting

1052
00:49:41,850 --> 00:49:44,370
the immediate impact
of the tenure decision,

1053
00:49:44,370 --> 00:49:47,460
but they overestimated
the long-run impact.

1054
00:49:47,460 --> 00:49:50,160
That is to say, in
fact, immediately

1055
00:49:50,160 --> 00:49:52,080
after being denied
tenure, people

1056
00:49:52,080 --> 00:49:54,810
are maybe arguably
or understandably

1057
00:49:54,810 --> 00:49:56,850
disappointed and unhappy.

1058
00:49:56,850 --> 00:49:59,850
But over time, the long run
effect is much less severe.

1059
00:49:59,850 --> 00:50:02,460
People seem to adjust
to their circumstances.

1060
00:50:02,460 --> 00:50:04,650
In the medium and
long run, people

1061
00:50:04,650 --> 00:50:08,940
are almost as happy
as if not as happy

1062
00:50:08,940 --> 00:50:11,220
as if they had gotten tenure.

1063
00:50:11,220 --> 00:50:13,650
So there's actually not
much of a long-run effect.

1064
00:50:13,650 --> 00:50:16,270
Yet people seem to
mispredict the effect.

1065
00:50:16,270 --> 00:50:18,190
They really seem to
think, in the short run,

1066
00:50:18,190 --> 00:50:19,440
the effect will be really bad.

1067
00:50:19,440 --> 00:50:23,020
And that bad effect
will last forever.

1068
00:50:23,020 --> 00:50:26,100
So people seem to essentially
sort of understand

1069
00:50:26,100 --> 00:50:30,600
there's a psychological
immune system to bad events.

1070
00:50:30,600 --> 00:50:34,830
People recover from negative
shocks quite well over time.

1071
00:50:34,830 --> 00:50:37,740
But people seem to mispredict
this recovery, which

1072
00:50:37,740 --> 00:50:41,010
which again is what's
called immune neglect.

1073
00:50:41,010 --> 00:50:43,830
There's similar misprediction
for other life events

1074
00:50:43,830 --> 00:50:47,370
such as paraplegia or
lottery wins, so for very

1075
00:50:47,370 --> 00:50:49,380
negative and positive events.

1076
00:50:49,380 --> 00:50:52,560
People seem to misunderstand
that, again, in the short run,

1077
00:50:52,560 --> 00:50:54,930
there's usually a
pretty large decrease

1078
00:50:54,930 --> 00:50:57,810
or increase in people's
happiness and life

1079
00:50:57,810 --> 00:50:58,920
satisfaction.

1080
00:50:58,920 --> 00:51:02,610
But people tend to recover
from that quite well over time.

1081
00:51:02,610 --> 00:51:04,380
But in their
prediction, people seem

1082
00:51:04,380 --> 00:51:06,840
to mispredict that adjustment.

1083
00:51:06,840 --> 00:51:11,580
People seem to think that the
bad effects on happiness--

1084
00:51:11,580 --> 00:51:13,920
or bad or good effects
and happiness--

1085
00:51:13,920 --> 00:51:16,420
would persist forever.

1086
00:51:16,420 --> 00:51:19,080
Let me now show you a simple
model of projection bias

1087
00:51:19,080 --> 00:51:20,790
by Loewenstein et al.

1088
00:51:20,790 --> 00:51:23,130
that formalizes some
of the intuitions

1089
00:51:23,130 --> 00:51:24,660
that we discussed so far.

1090
00:51:27,550 --> 00:51:29,640
So suppose true
utility at time t

1091
00:51:29,640 --> 00:51:32,190
depends on both
consumption ct at time t

1092
00:51:32,190 --> 00:51:34,120
and the state st at the time t.

1093
00:51:34,120 --> 00:51:37,850
So it's u of ct
st. The state could

1094
00:51:37,850 --> 00:51:40,910
be anything that affects utility
from consumption, ranging

1095
00:51:40,910 --> 00:51:42,920
from the level of
hunger or addiction,

1096
00:51:42,920 --> 00:51:45,050
whether somebody had
just smoked versus not

1097
00:51:45,050 --> 00:51:50,780
or whether somebody has just
gotten some BUP dose or not,

1098
00:51:50,780 --> 00:51:54,170
or past consumption, et cetera,
whether somebody is tired,

1099
00:51:54,170 --> 00:51:55,710
and so on.

1100
00:51:55,710 --> 00:52:01,190
Now, the prediction at time t
of future utility at the time

1101
00:52:01,190 --> 00:52:03,890
tower larger than t
from assuming the c tau

1102
00:52:03,890 --> 00:52:08,990
and state as tau is u
hat of c tau as tau.

1103
00:52:08,990 --> 00:52:12,200
And that equals 1
minus alpha times u

1104
00:52:12,200 --> 00:52:18,920
of c tau as tau plus
alpha times u of c tau st.

1105
00:52:18,920 --> 00:52:20,690
Now, what is this expression?

1106
00:52:20,690 --> 00:52:23,480
Well, if you look at the
first part, u of c tau

1107
00:52:23,480 --> 00:52:27,260
s tau, that is the
correct future utility

1108
00:52:27,260 --> 00:52:32,480
at time tau from consuming
c tau in state tau.

1109
00:52:32,480 --> 00:52:36,650
So if alpha was 0, then this
term here would go away,

1110
00:52:36,650 --> 00:52:38,360
this term is just 1.

1111
00:52:38,360 --> 00:52:42,840
So that is essentially
a correct prediction.

1112
00:52:42,840 --> 00:52:44,330
So there is no production bias.

1113
00:52:44,330 --> 00:52:46,400
The person has
rational expectation

1114
00:52:46,400 --> 00:52:53,480
about their future utility
at time tau of utility c

1115
00:52:53,480 --> 00:52:56,180
tau in state s tau.

1116
00:52:56,180 --> 00:52:59,210
Now, if alpha is not
zero, then there's

1117
00:52:59,210 --> 00:53:00,810
this additional term here.

1118
00:53:00,810 --> 00:53:04,318
So this is the weighted
average of alpha weights

1119
00:53:04,318 --> 00:53:06,110
for this other term
here, which we're going

1120
00:53:06,110 --> 00:53:07,193
to talk about in a second.

1121
00:53:07,193 --> 00:53:12,830
And 1 minus alpha, the
correct utility of consuming

1122
00:53:12,830 --> 00:53:15,020
c tau in state s tau.

1123
00:53:15,020 --> 00:53:16,770
Now, what is this
other term here?

1124
00:53:16,770 --> 00:53:19,640
Well, the person
gets c tau correctly,

1125
00:53:19,640 --> 00:53:22,370
so the consumption is right
but the state is wrong.

1126
00:53:22,370 --> 00:53:24,200
And what state is
the person using

1127
00:53:24,200 --> 00:53:25,910
to predict their future utility?

1128
00:53:25,910 --> 00:53:29,220
Well, it's using st,
the current state.

1129
00:53:29,220 --> 00:53:31,550
So the person is essentially
using their current state

1130
00:53:31,550 --> 00:53:34,370
to some degree to predict
their future utility instead

1131
00:53:34,370 --> 00:53:36,710
of the future state as tau.

1132
00:53:36,710 --> 00:53:39,702
Now, notice that if st
and s tau is the same,

1133
00:53:39,702 --> 00:53:40,910
then there's no problem here.

1134
00:53:40,910 --> 00:53:42,660
There's going to be
no misprediction.

1135
00:53:42,660 --> 00:53:45,080
So it better be the
case that the states st

1136
00:53:45,080 --> 00:53:46,760
and s tau are different.

1137
00:53:46,760 --> 00:53:51,860
For example, a hungry person
would mispredict that utility

1138
00:53:51,860 --> 00:53:53,600
from not being hungry.

1139
00:53:53,600 --> 00:53:56,860
So the utility from not being
hungry would be this part.

1140
00:53:56,860 --> 00:54:02,390
The person is hungry right now
at time t if st equals hungry.

1141
00:54:02,390 --> 00:54:04,928
The person would predict
the future utility even

1142
00:54:04,928 --> 00:54:07,220
when they're not hungry,
thinking that they will always

1143
00:54:07,220 --> 00:54:09,360
be hungry in the future.

1144
00:54:09,360 --> 00:54:13,390
So alpha in between 0 and 1 is
the degree of projection bias.

1145
00:54:17,970 --> 00:54:19,980
So just to give you
more detail, the person

1146
00:54:19,980 --> 00:54:22,460
predicts how she'd
feel about consuming c

1147
00:54:22,460 --> 00:54:26,010
tau in the future partially
at least by how she'd

1148
00:54:26,010 --> 00:54:27,930
feel about consuming it now.

1149
00:54:27,930 --> 00:54:30,240
That's what the parameter
alpha is measuring.

1150
00:54:30,240 --> 00:54:32,310
Alpha between 0 and
1-- it could be 0.

1151
00:54:32,310 --> 00:54:33,060
It could be 1--

1152
00:54:33,060 --> 00:54:35,050
is the degree of
projection bias.

1153
00:54:35,050 --> 00:54:37,510
So alpha equals 0 is
correct understanding

1154
00:54:37,510 --> 00:54:38,520
of future utility.

1155
00:54:38,520 --> 00:54:40,260
That's essentially
no projection bias.

1156
00:54:40,260 --> 00:54:42,180
That's rational expectations.

1157
00:54:42,180 --> 00:54:44,533
Alpha equals 1 is
full projection bias.

1158
00:54:44,533 --> 00:54:46,200
That's a person who
thinks, essentially,

1159
00:54:46,200 --> 00:54:48,165
their future state
will always be

1160
00:54:48,165 --> 00:54:50,040
their current state,
which, of course, is not

1161
00:54:50,040 --> 00:54:51,865
necessarily true.

1162
00:54:51,865 --> 00:54:54,240
So now that person will optimize
according to a perceived

1163
00:54:54,240 --> 00:54:55,980
future preferences.

1164
00:54:55,980 --> 00:54:59,640
You had c tau, s
tau, and I'm going

1165
00:54:59,640 --> 00:55:01,680
to assume for now is an
exponential discounter.

1166
00:55:01,680 --> 00:55:04,710
That's, of course, you
could relax that easily.

1167
00:55:04,710 --> 00:55:06,210
And, in fact, in
some of the problem

1168
00:55:06,210 --> 00:55:08,760
sets from previous years
and from this year,

1169
00:55:08,760 --> 00:55:11,760
that will be, in fact, relaxed.

1170
00:55:11,760 --> 00:55:14,580
Now let me give you a very
simple example with hunger.

1171
00:55:14,580 --> 00:55:19,830
Again, the problem set this
year and past problem sets

1172
00:55:19,830 --> 00:55:21,960
give you quite a few
other additional examples

1173
00:55:21,960 --> 00:55:23,220
that you can study.

1174
00:55:23,220 --> 00:55:25,370
So suppose there are two states.

1175
00:55:25,370 --> 00:55:28,505
The state is either hungry
or not hungry, H or N.

1176
00:55:28,505 --> 00:55:32,610
And the consumption ct is
over burgers and money.

1177
00:55:32,610 --> 00:55:37,680
So u of ct where ct is the
number of burgers and money

1178
00:55:37,680 --> 00:55:40,260
that the person has.

1179
00:55:40,260 --> 00:55:44,213
And state H is 5 times
the number of burgers

1180
00:55:44,213 --> 00:55:46,380
that the person eats plus
the remaining money that's

1181
00:55:46,380 --> 00:55:48,660
left to spend on other things.

1182
00:55:48,660 --> 00:55:51,480
The utility of ct and when
the person is not hungry

1183
00:55:51,480 --> 00:55:56,820
is 1 times the number of burgers
plus the money that's left.

1184
00:55:56,820 --> 00:55:58,590
Now what does that imply?

1185
00:55:58,590 --> 00:56:01,770
Well, it implies that she's
willing to pay $5 for a burger

1186
00:56:01,770 --> 00:56:04,260
when hungry and $1 when full.

1187
00:56:04,260 --> 00:56:05,560
How do we know that?

1188
00:56:05,560 --> 00:56:08,670
Well, we know that because
in the first case when

1189
00:56:08,670 --> 00:56:12,990
the person is hungry,
each burger gives 5 utils,

1190
00:56:12,990 --> 00:56:14,610
and each dollar gets 1 util.

1191
00:56:14,610 --> 00:56:18,270
So the exchange rate between
burgers and utils is 5.

1192
00:56:18,270 --> 00:56:21,540
So the person is going
to pay $5 per burger.

1193
00:56:21,540 --> 00:56:23,190
In the second case,
the exchange trade

1194
00:56:23,190 --> 00:56:27,360
between utility from burgers
and utility from dollars is 1.

1195
00:56:27,360 --> 00:56:31,330
So the person is willing
to pay $1 for a burger.

1196
00:56:31,330 --> 00:56:33,670
Now suppose alpha is 3/4.

1197
00:56:33,670 --> 00:56:36,273
Again, remember alpha is the
degree of projection bias.

1198
00:56:36,273 --> 00:56:37,690
I'm going to go
back and show you.

1199
00:56:37,690 --> 00:56:39,340
Alpha is this parameter here.

1200
00:56:39,340 --> 00:56:41,230
Alpha is the degree
of projection bias,

1201
00:56:41,230 --> 00:56:43,550
so the degree of misprediction.

1202
00:56:43,550 --> 00:56:47,920
So suppose in our example
here alpha equals 3/4,

1203
00:56:47,920 --> 00:56:51,325
and the person is
not hungry right now.

1204
00:56:51,325 --> 00:56:53,200
What is their willingness
to pay for a burger

1205
00:56:53,200 --> 00:56:55,000
tonight when she'll be hungry?

1206
00:56:55,000 --> 00:56:57,280
So if the person is
not hungry right now,

1207
00:56:57,280 --> 00:57:00,040
her correct utility
or willingness

1208
00:57:00,040 --> 00:57:01,900
to pay for a burger is $1.

1209
00:57:01,900 --> 00:57:03,340
We just established
that already.

1210
00:57:03,340 --> 00:57:04,360
She's full.

1211
00:57:04,360 --> 00:57:08,470
And the correct utility for
willingness to pay for a burger

1212
00:57:08,470 --> 00:57:12,440
tonight when she'll
be hungry is $5.

1213
00:57:12,440 --> 00:57:15,630
Now her projection bias is 3/4.

1214
00:57:15,630 --> 00:57:22,280
So essentially she puts 3/4 of
a weight on her current state

1215
00:57:22,280 --> 00:57:24,680
and 1/4 on the full state.

1216
00:57:24,680 --> 00:57:28,040
Well, 3/4 times 1
plus 1/4 times 5

1217
00:57:28,040 --> 00:57:31,145
or going essentially 3/4
between a difference of 4

1218
00:57:31,145 --> 00:57:34,130
is essentially 3.

1219
00:57:34,130 --> 00:57:35,810
So the person
essentially will think

1220
00:57:35,810 --> 00:57:43,460
that their future willingness to
pay is $2 for a burger tonight.

1221
00:57:43,460 --> 00:57:48,470
Or put differently, if she
thinks she only wants like--

1222
00:57:48,470 --> 00:57:51,410
her willingness to pay will
only be $2 for a burger tonight,

1223
00:57:51,410 --> 00:57:52,910
if she orders for
a burger tonight

1224
00:57:52,910 --> 00:57:55,400
and the burger
costs like $3 or $4,

1225
00:57:55,400 --> 00:57:58,860
she will not be
willing to do that.

1226
00:57:58,860 --> 00:58:04,220
Now just to be very clear, so
the person only essentially--

1227
00:58:04,220 --> 00:58:07,850
if alpha equals
3/4, the person only

1228
00:58:07,850 --> 00:58:10,580
incorporates 1/4 of
the preference change

1229
00:58:10,580 --> 00:58:14,190
from being full versus
hungry in the future.

1230
00:58:14,190 --> 00:58:18,845
So that is to say there's a
difference of $4 between $1

1231
00:58:18,845 --> 00:58:23,780
and $5, and she only takes into
account 1/4 of that $4, which

1232
00:58:23,780 --> 00:58:24,740
is $1.

1233
00:58:24,740 --> 00:58:27,980
So willingness to pay
is 1 plus 1 equals 2.

1234
00:58:27,980 --> 00:58:29,780
You can also write
this down here

1235
00:58:29,780 --> 00:58:32,930
instead of using the
formula from the equation

1236
00:58:32,930 --> 00:58:37,250
from before, which essentially
is 3/4, which is, again,

1237
00:58:37,250 --> 00:58:43,310
alpha times utility
using the current state,

1238
00:58:43,310 --> 00:58:47,900
and 1/4 using the
correct utility, which

1239
00:58:47,900 --> 00:58:48,920
is the future state.

1240
00:58:48,920 --> 00:58:51,140
Once you sort of
calculate that, she

1241
00:58:51,140 --> 00:58:54,380
would be willing to pay
at most $2 for a burger.

1242
00:58:54,380 --> 00:58:56,360
So potentially she's
making a mistake.

1243
00:59:00,050 --> 00:59:06,600
And so again in
the problem set you

1244
00:59:06,600 --> 00:59:09,120
will have more
examples and questions

1245
00:59:09,120 --> 00:59:12,540
that allows you to practice
with this level model.

1246
00:59:12,540 --> 00:59:15,400
Now, why should you care
about projection bias?

1247
00:59:15,400 --> 00:59:16,860
I showed you pretty
good evidence

1248
00:59:16,860 --> 00:59:21,320
of projection bias in many
settings in the sense of saying

1249
00:59:21,320 --> 00:59:24,510
we think projection bias really
exists and we think projection

1250
00:59:24,510 --> 00:59:26,410
bias might be quite important.

1251
00:59:26,410 --> 00:59:26,910
Sorry.

1252
00:59:26,910 --> 00:59:30,210
It might be relevant in
explaining people's behavior

1253
00:59:30,210 --> 00:59:33,020
in those settings.

1254
00:59:33,020 --> 00:59:35,930
Now, the consequences
in those settings

1255
00:59:35,930 --> 00:59:37,790
is relatively unimportant.

1256
00:59:37,790 --> 00:59:41,150
For example, in car purchases
if rich people buy convertibles

1257
00:59:41,150 --> 00:59:43,940
and then sort of return
their convertibles,

1258
00:59:43,940 --> 00:59:46,040
the loss from that is
not particularly large.

1259
00:59:46,040 --> 00:59:49,940
Maybe there's some
redistribution across people.

1260
00:59:49,940 --> 00:59:51,860
Similarly, the
catalog orders, people

1261
00:59:51,860 --> 00:59:55,040
might purchase too many
things and return them.

1262
00:59:55,040 --> 00:59:58,100
Really, the consequences
are not that large,

1263
00:59:58,100 --> 01:00:01,010
and the actions are
often reversible.

1264
01:00:01,010 --> 01:00:03,702
So why should we then care
about projection bias?

1265
01:00:03,702 --> 01:00:06,035
Well, there's at least two
really important applications

1266
01:00:06,035 --> 01:00:09,380
that we're going to talk about
for a little bit that really

1267
01:00:09,380 --> 01:00:10,820
could matter quite a lot.

1268
01:00:10,820 --> 01:00:12,470
One is addiction,
which is people

1269
01:00:12,470 --> 01:00:15,140
mispredict the
utility from becoming

1270
01:00:15,140 --> 01:00:17,310
addicted in the future.

1271
01:00:17,310 --> 01:00:18,818
Second is depression and hope.

1272
01:00:18,818 --> 01:00:20,360
When people are
depressed, they might

1273
01:00:20,360 --> 01:00:24,650
mispredict how they feel when
they're not depressed anymore.

1274
01:00:24,650 --> 01:00:29,930
Some might argue that
marriage and relationships are

1275
01:00:29,930 --> 01:00:32,430
quite important or projection
bias might be quite important.

1276
01:00:32,430 --> 01:00:34,130
Particularly when
people have fights,

1277
01:00:34,130 --> 01:00:36,350
they might want to get
divorced very quickly.

1278
01:00:36,350 --> 01:00:38,630
They might sort
of do rash things

1279
01:00:38,630 --> 01:00:40,190
about getting married
really quickly

1280
01:00:40,190 --> 01:00:42,550
or try to get a
divorce very quickly.

1281
01:00:42,550 --> 01:00:45,040
And, in fact, there are some
laws that often prevent that.

1282
01:00:45,040 --> 01:00:47,540
People need to actually have
some cool-down period for a few

1283
01:00:47,540 --> 01:00:50,330
days where they
have to sort of--

1284
01:00:50,330 --> 01:00:52,160
if one wants to get
married, for example,

1285
01:00:52,160 --> 01:00:54,285
you have to get a license,
and it takes a few days,

1286
01:00:54,285 --> 01:00:56,240
and only then you
can get married.

1287
01:00:56,240 --> 01:00:59,360
The same is true for a divorce,
at least in some places.

1288
01:00:59,360 --> 01:01:03,950
Perhaps that's because people
suffer from projection bias

1289
01:01:03,950 --> 01:01:04,520
sometimes.

1290
01:01:04,520 --> 01:01:06,740
And when they're really mad
at somebody or something

1291
01:01:06,740 --> 01:01:09,115
really bad happens between
them or something really great

1292
01:01:09,115 --> 01:01:10,940
happened between
them, they might not

1293
01:01:10,940 --> 01:01:14,780
be able to predict that their
preferences and their views

1294
01:01:14,780 --> 01:01:19,543
towards the other person
might change over time.

1295
01:01:19,543 --> 01:01:21,710
But let me tell you now a
little bit about addiction

1296
01:01:21,710 --> 01:01:24,663
and depression and hope.

1297
01:01:24,663 --> 01:01:26,830
So first addiction, and
that's the main applications

1298
01:01:26,830 --> 01:01:28,520
we're going to discuss.

1299
01:01:28,520 --> 01:01:31,460
So projection bias might
be important for people's

1300
01:01:31,460 --> 01:01:36,870
initiation when people are
starting to take drugs.

1301
01:01:36,870 --> 01:01:38,870
So here let's define
the relevant state

1302
01:01:38,870 --> 01:01:40,760
as the person's
level of addiction.

1303
01:01:40,760 --> 01:01:43,820
So how much has the person
consumed of cigarettes

1304
01:01:43,820 --> 01:01:46,070
in the last month or year?

1305
01:01:46,070 --> 01:01:48,440
In an unaddicted state,
cigarettes are really not

1306
01:01:48,440 --> 01:01:50,240
that hard to resist, right?

1307
01:01:50,240 --> 01:01:52,370
If somebody has
never smoked before,

1308
01:01:52,370 --> 01:01:54,140
cigarettes are really
not that attractive.

1309
01:01:54,140 --> 01:01:56,720
Same is true also for
alcohol, for example.

1310
01:01:56,720 --> 01:01:59,075
In an unaddicted-- so
in an addicted state,

1311
01:01:59,075 --> 01:02:00,950
however, it's very hard
to resist cigarettes.

1312
01:02:00,950 --> 01:02:02,367
If you have smoked
a lot recently,

1313
01:02:02,367 --> 01:02:05,450
you really, really would like
to smoke on any given day.

1314
01:02:05,450 --> 01:02:08,540
Now, the unaddicted
person might sort of

1315
01:02:08,540 --> 01:02:11,330
think that experimenting
with cigarettes is fun

1316
01:02:11,330 --> 01:02:12,740
but does not want
to get addicted

1317
01:02:12,740 --> 01:02:13,980
for the rest of her life.

1318
01:02:13,980 --> 01:02:15,260
It's sort of fun to
do it for a while,

1319
01:02:15,260 --> 01:02:17,177
but really then you want
to stop because being

1320
01:02:17,177 --> 01:02:19,850
addicted for the rest of
your life is really bad.

1321
01:02:19,850 --> 01:02:23,840
You might get really bad health
consequences, for instance.

1322
01:02:23,840 --> 01:02:25,610
And that might give
you some false--

1323
01:02:25,610 --> 01:02:27,710
if you sort of have
projection bias,

1324
01:02:27,710 --> 01:02:30,260
you might have some
false sense of control.

1325
01:02:30,260 --> 01:02:32,930
If you project your current
nonaddictive preferences

1326
01:02:32,930 --> 01:02:36,380
into the future, you might
think that you can stop smoking

1327
01:02:36,380 --> 01:02:38,700
or taking drugs if necessary.

1328
01:02:38,700 --> 01:02:41,810
And so then the person might
try cigarettes, get addicted,

1329
01:02:41,810 --> 01:02:43,430
and consume much of your life.

1330
01:02:43,430 --> 01:02:46,700
Perhaps she wouldn't do so
if she knew she could quit.

1331
01:02:46,700 --> 01:02:49,340
That's essentially to say if
you're not addicted right now,

1332
01:02:49,340 --> 01:02:51,140
you might predict
that it's going to be

1333
01:02:51,140 --> 01:02:52,650
always easy to stop smoking.

1334
01:02:52,650 --> 01:02:55,550
And, in fact, right
now it is really easy

1335
01:02:55,550 --> 01:02:58,370
to stop smoking or resisting.

1336
01:02:58,370 --> 01:03:02,270
But the person might mispredict
that this will always

1337
01:03:02,270 --> 01:03:05,870
be the case in the future
even once she is addicted

1338
01:03:05,870 --> 01:03:08,360
and once she has
smoked a lot because

1339
01:03:08,360 --> 01:03:11,990
for the nonaddicted person,
it's really hard to imagine

1340
01:03:11,990 --> 01:03:15,760
that it might be even harder
to resist in the future.

1341
01:03:15,760 --> 01:03:20,010
In addition, there are
quitting and restarting

1342
01:03:20,010 --> 01:03:22,960
cycles that are very common
when it comes to addiction.

1343
01:03:22,960 --> 01:03:24,840
So addicts often
express a desire

1344
01:03:24,840 --> 01:03:27,120
to stop using
substances permanently

1345
01:03:27,120 --> 01:03:29,010
but are unable to
follow through.

1346
01:03:29,010 --> 01:03:31,170
That's not surprising,
and we've kind of

1347
01:03:31,170 --> 01:03:32,760
thought about this before.

1348
01:03:32,760 --> 01:03:34,890
And quasi hyperbolic
discounting can essentially

1349
01:03:34,890 --> 01:03:36,960
predict that as well.

1350
01:03:39,720 --> 01:03:41,550
Short-term abstention
in addiction

1351
01:03:41,550 --> 01:03:45,430
is common while long-term
abstention is rare.

1352
01:03:45,430 --> 01:03:48,150
So in 2000, for
example, 41% of smokers

1353
01:03:48,150 --> 01:03:53,910
stopped for at least one day
trying to quit, but only 4.7%

1354
01:03:53,910 --> 01:03:56,640
successfully abstained for
more than three months.

1355
01:03:56,640 --> 01:04:00,250
That doesn't really seem like
quasi hyperbolic discounting.

1356
01:04:00,250 --> 01:04:04,200
So you wouldn't go through
a pointless short-term pain

1357
01:04:04,200 --> 01:04:06,300
and then not follow
through if [INAUDIBLE]

1358
01:04:06,300 --> 01:04:08,070
quasi hyperbolic discounting.

1359
01:04:08,070 --> 01:04:11,760
The reason being in particular
that withdrawal symptoms

1360
01:04:11,760 --> 01:04:13,590
tend to be, on
average, strongest

1361
01:04:13,590 --> 01:04:15,660
at the start of a quit attempt.

1362
01:04:15,660 --> 01:04:17,520
So hyperbolic discounting
would not sort of

1363
01:04:17,520 --> 01:04:24,183
go over this really difficult
pain in the short run

1364
01:04:24,183 --> 01:04:26,350
if you don't follow through
because it's essentially

1365
01:04:26,350 --> 01:04:27,352
short-run pain.

1366
01:04:30,020 --> 01:04:33,080
And then three, recidivism
rates are especially high when

1367
01:04:33,080 --> 01:04:36,470
addicts are exposed to
occasional cues related

1368
01:04:36,470 --> 01:04:38,130
to past consumption.

1369
01:04:38,130 --> 01:04:41,480
So treatment programs, in
fact, advise recovering addicts

1370
01:04:41,480 --> 01:04:44,360
to move to new locations
and to avoid places

1371
01:04:44,360 --> 01:04:49,080
when previous
consumption took place.

1372
01:04:49,080 --> 01:04:51,080
So let's sort of go
through these one by one

1373
01:04:51,080 --> 01:04:53,330
and try to see whether,
particularly in two and three,

1374
01:04:53,330 --> 01:04:58,390
whether we can explain
those using projection bias.

1375
01:04:58,390 --> 01:05:01,800
So now let's define the state
as the strength of cravings

1376
01:05:01,800 --> 01:05:04,020
at the moment, and suppose
this varies randomly

1377
01:05:04,020 --> 01:05:06,090
or with exposure to cues.

1378
01:05:06,090 --> 01:05:09,570
How do you explain the
starting a quit attempt?

1379
01:05:09,570 --> 01:05:12,690
Well, suppose an addict is
currently consuming regularly.

1380
01:05:12,690 --> 01:05:14,528
She experiences
periods of low cravings

1381
01:05:14,528 --> 01:05:15,570
when it's easy to resist.

1382
01:05:15,570 --> 01:05:19,110
That's like when the person has
just smoked, cravings are low,

1383
01:05:19,110 --> 01:05:20,610
and it's easy to resist.

1384
01:05:20,610 --> 01:05:22,650
So then the person
might think, well, it'll

1385
01:05:22,650 --> 01:05:24,750
always be easy to
resist, so she thinks

1386
01:05:24,750 --> 01:05:28,770
it's worth trying to quit and
starts the quitting attempt.

1387
01:05:28,770 --> 01:05:30,960
Now, how do we explain
then how people

1388
01:05:30,960 --> 01:05:32,550
abandon the quit attempt?

1389
01:05:32,550 --> 01:05:35,100
Well, suppose the addict is
currently on a quit attempt.

1390
01:05:35,100 --> 01:05:38,160
Something triggers strong
cravings, which might just

1391
01:05:38,160 --> 01:05:40,470
be over time you get
cravings or maybe the person

1392
01:05:40,470 --> 01:05:43,410
might just be
exposed to some cues.

1393
01:05:43,410 --> 01:05:45,360
So then the person
feels it's really hard

1394
01:05:45,360 --> 01:05:48,060
to resist the drugs,
and in particular, she

1395
01:05:48,060 --> 01:05:51,625
thinks drugs will always be
hard to resist because currently

1396
01:05:51,625 --> 01:05:53,250
she's craving a lot,
and you can't even

1397
01:05:53,250 --> 01:05:56,030
imagine that it might be--

1398
01:05:56,030 --> 01:05:58,270
that that craving
might stop over time.

1399
01:05:58,270 --> 01:06:01,020
So she thinks the quit attempt
is impossible to carry through,

1400
01:06:01,020 --> 01:06:02,520
and so she abandons it.

1401
01:06:02,520 --> 01:06:05,280
So that way projection
bias can both explain

1402
01:06:05,280 --> 01:06:07,410
why the person starts
the quit attempt

1403
01:06:07,410 --> 01:06:09,840
but also why the person
does not follow through.

1404
01:06:16,578 --> 01:06:17,526
I'm sorry.

1405
01:06:17,526 --> 01:06:19,620
Give me a second.

1406
01:06:19,620 --> 01:06:21,990
How do we think
about the recidivism?

1407
01:06:21,990 --> 01:06:25,545
Well, here there's
not necessarily

1408
01:06:25,545 --> 01:06:27,180
a projection bias needed.

1409
01:06:27,180 --> 01:06:31,500
But here people might
misunderestimate

1410
01:06:31,500 --> 01:06:36,520
how important cues are in
affecting people's utility.

1411
01:06:36,520 --> 01:06:38,760
So if it's a case that an
addict, for example-- when

1412
01:06:38,760 --> 01:06:40,920
the addict walks by,
like, a liquor store

1413
01:06:40,920 --> 01:06:43,890
or when the addict
is talking to friends

1414
01:06:43,890 --> 01:06:47,400
who they have been taking
drugs in the past and so on,

1415
01:06:47,400 --> 01:06:50,400
that might really increase
people's marginal utility

1416
01:06:50,400 --> 01:06:51,730
of using drugs.

1417
01:06:51,730 --> 01:06:53,790
So they might get really,
really strong cravings,

1418
01:06:53,790 --> 01:06:56,580
and it might be really, really
hard for them to resist.

1419
01:06:56,580 --> 01:07:02,040
Now if you know that as an
addict who is recovering,

1420
01:07:02,040 --> 01:07:03,870
who's not taking
any drugs currently,

1421
01:07:03,870 --> 01:07:06,090
you should at all
cost avoid these cues

1422
01:07:06,090 --> 01:07:08,430
because you know you want
to make sure that you're not

1423
01:07:08,430 --> 01:07:10,440
exposed to them.

1424
01:07:10,440 --> 01:07:14,070
But if you have protection
bias, you might underestimate

1425
01:07:14,070 --> 01:07:17,400
the importance of
those cues in affecting

1426
01:07:17,400 --> 01:07:20,520
your utility or your
cravings, and so

1427
01:07:20,520 --> 01:07:22,710
you might sort of walk
into a bar thinking like,

1428
01:07:22,710 --> 01:07:24,668
well, I haven't really
been smoking or drinking

1429
01:07:24,668 --> 01:07:27,210
for a long time, so I'm sure
they will be able to resist.

1430
01:07:27,210 --> 01:07:30,150
But, of course, then once
you're exposed to these cues,

1431
01:07:30,150 --> 01:07:32,620
you will not be able to do
so, and that misprediction

1432
01:07:32,620 --> 01:07:34,860
might really then
cause recidivism.

1433
01:07:34,860 --> 01:07:37,770
So again, in that
sense, protection bias

1434
01:07:37,770 --> 01:07:42,760
is also consistent with that
kind of recidivism due to cues.

1435
01:07:42,760 --> 01:07:48,478
Now let me talk very briefly
only about depression.

1436
01:07:48,478 --> 01:07:50,020
And the reason being
that we're going

1437
01:07:50,020 --> 01:07:52,280
to talk about mental
health in a future lecture.

1438
01:07:52,280 --> 01:07:54,160
So depressed individuals
have the tendency

1439
01:07:54,160 --> 01:07:57,560
to project the depressed
feelings not only to the future

1440
01:07:57,560 --> 01:07:59,420
but also to the past.

1441
01:07:59,420 --> 01:08:05,380
And so in particular, the
depressed tend to think about--

1442
01:08:05,380 --> 01:08:08,605
depression is the inability
to construct a future.

1443
01:08:08,605 --> 01:08:10,480
And in particular, when
someone is depressed,

1444
01:08:10,480 --> 01:08:13,780
the past and future are absorbed
entirely by the present,

1445
01:08:13,780 --> 01:08:15,910
and people can neither
remember feeling better

1446
01:08:15,910 --> 01:08:19,553
nor imagine that they will
feel better in the future.

1447
01:08:19,553 --> 01:08:21,220
So that's to say if
the person is really

1448
01:08:21,220 --> 01:08:24,370
feeling terrible
right now, they also

1449
01:08:24,370 --> 01:08:26,920
think they always have been
feeling terrible in the past,

1450
01:08:26,920 --> 01:08:28,630
and they also think
that they will always

1451
01:08:28,630 --> 01:08:30,649
feel terrible in the future.

1452
01:08:30,649 --> 01:08:33,220
So then life might feel
particularly hopeful

1453
01:08:33,220 --> 01:08:35,350
if there's no scope for
future improvements.

1454
01:08:35,350 --> 01:08:37,700
And if you think your
life was always that bad,

1455
01:08:37,700 --> 01:08:39,979
there's really--

1456
01:08:39,979 --> 01:08:41,979
in some cases, people
might think there's really

1457
01:08:41,979 --> 01:08:44,260
not much of a reason to live.

1458
01:08:44,260 --> 01:08:46,420
And that's, of course,
an extremely dangerous

1459
01:08:46,420 --> 01:08:54,490
and difficult or consequential,
potentially, misprediction

1460
01:08:54,490 --> 01:08:56,200
because, of course,
depressed people

1461
01:08:56,200 --> 01:08:59,859
can get better through therapy
and through drugs and so on.

1462
01:09:02,380 --> 01:09:05,319
And often there's also
remission over time.

1463
01:09:05,319 --> 01:09:09,167
But if people mispredict that,
they might, for instance, never

1464
01:09:09,167 --> 01:09:11,500
seek treatment because they
think, why would I ever seek

1465
01:09:11,500 --> 01:09:13,840
treatment if I'm
always feeling terrible

1466
01:09:13,840 --> 01:09:15,757
for the rest of my life anyway?

1467
01:09:15,757 --> 01:09:17,590
We're going to get back
to that a little bit

1468
01:09:17,590 --> 01:09:20,899
when talking about
mental health.

1469
01:09:20,899 --> 01:09:28,024
Now, one thing that we have
not really talked about

1470
01:09:28,024 --> 01:09:29,899
is how to think about
these different biases.

1471
01:09:29,899 --> 01:09:31,910
In particular,
what's the difference

1472
01:09:31,910 --> 01:09:34,700
between projection bias
and naive quasi-hyperbolic

1473
01:09:34,700 --> 01:09:36,029
discounting?

1474
01:09:36,029 --> 01:09:38,660
So both naivety and
quasi-hyperbolic discounting

1475
01:09:38,660 --> 01:09:41,609
and projection bias
entail a misprediction.

1476
01:09:41,609 --> 01:09:44,840
So as an example, take a
smoker who wants to quit.

1477
01:09:44,840 --> 01:09:47,630
The naive quasi-hyperbolic
discounter

1478
01:09:47,630 --> 01:09:50,120
overestimates their
future patience.

1479
01:09:50,120 --> 01:09:53,240
So that person will sign up
for a commitment contract

1480
01:09:53,240 --> 01:09:56,540
to stop smoking and might
fail due to the overestimation

1481
01:09:56,540 --> 01:09:58,460
of usefulness of
commitment devices.

1482
01:09:58,460 --> 01:10:02,030
That person thinks
that commitments--

1483
01:10:02,030 --> 01:10:05,060
the person understands
that she has some--

1484
01:10:05,060 --> 01:10:07,683
suppose that's a partially
naive person, which we need.

1485
01:10:07,683 --> 01:10:09,350
Otherwise the person
would never sign up

1486
01:10:09,350 --> 01:10:10,700
for a commitment contract.

1487
01:10:10,700 --> 01:10:13,220
Suppose the person
is partially naive.

1488
01:10:13,220 --> 01:10:15,590
The person will sign up for
that commitment contract

1489
01:10:15,590 --> 01:10:18,560
thinking that's going to
help them stop smoking,

1490
01:10:18,560 --> 01:10:20,300
but the person is
going to underestimate

1491
01:10:20,300 --> 01:10:22,550
how bad their
self-control problems are.

1492
01:10:22,550 --> 01:10:24,470
So then the person
might fail because they

1493
01:10:24,470 --> 01:10:27,012
think their commitment device
is really, really helpful when,

1494
01:10:27,012 --> 01:10:29,450
in fact, it's not strong enough.

1495
01:10:29,450 --> 01:10:33,540
Now for projection
bias, the person

1496
01:10:33,540 --> 01:10:35,220
might underestimate
the influence

1497
01:10:35,220 --> 01:10:37,653
of altered future states.

1498
01:10:37,653 --> 01:10:40,070
So the person might also sign
up for a commitment contract

1499
01:10:40,070 --> 01:10:42,300
to stop smoking
and then might feel

1500
01:10:42,300 --> 01:10:45,420
guilty due to the
underestimation of the changes

1501
01:10:45,420 --> 01:10:46,890
in future cravings.

1502
01:10:46,890 --> 01:10:49,673
That is to say if you are
currently not craving,

1503
01:10:49,673 --> 01:10:51,090
you might sign up
for a commitment

1504
01:10:51,090 --> 01:10:56,100
contract that would help you
to continue not using drugs.

1505
01:10:56,100 --> 01:10:59,130
And you might
underestimate, however,

1506
01:10:59,130 --> 01:11:01,500
how bad the cravings
will be in the future

1507
01:11:01,500 --> 01:11:03,640
once you see certain
cues or the like.

1508
01:11:03,640 --> 01:11:06,360
So again, you might choose
a commitment contract

1509
01:11:06,360 --> 01:11:09,060
that is not strong enough
because you mispredict

1510
01:11:09,060 --> 01:11:10,410
your future preferences.

1511
01:11:10,410 --> 01:11:12,630
In this case, not because
it's in the future,

1512
01:11:12,630 --> 01:11:15,690
not because you mispredicting
beta like in case number one

1513
01:11:15,690 --> 01:11:17,640
but rather because
you mispredict

1514
01:11:17,640 --> 01:11:23,333
how strongly future states
will affect your preferences.

1515
01:11:23,333 --> 01:11:25,500
You can also think about
another example, anesthesia

1516
01:11:25,500 --> 01:11:26,550
during childbirth.

1517
01:11:26,550 --> 01:11:31,470
You can get similar
sort of patterns.

1518
01:11:31,470 --> 01:11:35,010
Both naive quasi-hyperbolic
discounting and projection bias

1519
01:11:35,010 --> 01:11:37,875
might be able to explain
people's choices there.

1520
01:11:37,875 --> 01:11:40,830
In particular, both of
them would potentially

1521
01:11:40,830 --> 01:11:44,120
predict preference reversals.

1522
01:11:44,120 --> 01:11:48,810
Now, how can we tell projection
bias and naivety regarding beta

1523
01:11:48,810 --> 01:11:50,090
apart?

1524
01:11:50,090 --> 01:11:52,610
Well, the key part here
is that projection bias

1525
01:11:52,610 --> 01:11:55,140
is a state-dependent
misprediction.

1526
01:11:55,140 --> 01:11:57,530
So people are more likely
to predict future temptation

1527
01:11:57,530 --> 01:11:59,270
to overeat when hungry.

1528
01:11:59,270 --> 01:12:01,010
They're more likely
to predict smoking

1529
01:12:01,010 --> 01:12:03,350
when you haven't had a
cigarette for a while.

1530
01:12:03,350 --> 01:12:06,110
So when people are-- when
the state in the future

1531
01:12:06,110 --> 01:12:07,940
is different from
the current state,

1532
01:12:07,940 --> 01:12:11,040
people might mispredict
their preferences.

1533
01:12:11,040 --> 01:12:15,320
So now a present
bias, in fact, has

1534
01:12:15,320 --> 01:12:17,750
nothing to do with the
state, essentially just about

1535
01:12:17,750 --> 01:12:20,630
the future versus the present.

1536
01:12:20,630 --> 01:12:22,180
So now what do we need?

1537
01:12:22,180 --> 01:12:25,640
What kind of variation do we
need to disentangle the two

1538
01:12:25,640 --> 01:12:26,630
explanations?

1539
01:12:26,630 --> 01:12:28,670
Well, you need
variation in timing

1540
01:12:28,670 --> 01:12:31,580
and in particular
variation in states.

1541
01:12:31,580 --> 01:12:34,433
So present bias would
say it doesn't really

1542
01:12:34,433 --> 01:12:35,600
matter what state you're in.

1543
01:12:35,600 --> 01:12:38,110
It doesn't matter whether
you're hungry or not right now.

1544
01:12:38,110 --> 01:12:39,860
There's always going
to be a misprediction

1545
01:12:39,860 --> 01:12:42,510
between the present
and the future.

1546
01:12:42,510 --> 01:12:46,030
And projection bias would very
much say the states matter.

1547
01:12:46,030 --> 01:12:49,180
It's a state-dependent
misprediction.

1548
01:12:49,180 --> 01:12:54,390
So if you wanted to look
at people's predictions

1549
01:12:54,390 --> 01:12:58,470
for the future, present bias or
naivete regarding present bias

1550
01:12:58,470 --> 01:13:00,780
would say it doesn't matter
whether you have smoked

1551
01:13:00,780 --> 01:13:03,030
or whether you're hungry
right now for your prediction

1552
01:13:03,030 --> 01:13:04,050
for the future.

1553
01:13:04,050 --> 01:13:07,770
Projection bias would say
it does matter quite a bit.

1554
01:13:07,770 --> 01:13:10,830
But then a tricky question
is when should you,

1555
01:13:10,830 --> 01:13:13,470
in fact, offer people
commitment devices?

1556
01:13:13,470 --> 01:13:16,410
Well, you should probably
offer people commitment devices

1557
01:13:16,410 --> 01:13:18,720
if you want them
to choose correct--

1558
01:13:24,310 --> 01:13:26,760
if you want people to
make correct choices.

1559
01:13:26,760 --> 01:13:30,090
You probably want
to offer people

1560
01:13:30,090 --> 01:13:32,070
commitment devices at
times when they're,

1561
01:13:32,070 --> 01:13:34,000
in fact, in the same state.

1562
01:13:34,000 --> 01:13:36,200
So when somebody
already has cravings,

1563
01:13:36,200 --> 01:13:37,950
you want to offer them
a commitment device

1564
01:13:37,950 --> 01:13:39,533
because that person
really understands

1565
01:13:39,533 --> 01:13:41,490
how it will feel in the future.

1566
01:13:41,490 --> 01:13:44,280
If, instead, you offer somebody
a commitment device when

1567
01:13:44,280 --> 01:13:46,260
they don't have
cravings, they might

1568
01:13:46,260 --> 01:13:48,810
choose that commitment
device, which I guess is good.

1569
01:13:48,810 --> 01:13:51,600
But then they might fail
because they mispredict

1570
01:13:51,600 --> 01:13:53,580
how strong of a
commitment device

1571
01:13:53,580 --> 01:13:55,140
they might need in the future.

1572
01:13:58,270 --> 01:14:00,167
OK, so let me then
briefly summarize.

1573
01:14:00,167 --> 01:14:01,250
So what did we talk about?

1574
01:14:01,250 --> 01:14:03,370
So we talked about
state-dependent preferences.

1575
01:14:03,370 --> 01:14:05,200
So preferences
vary systematically

1576
01:14:05,200 --> 01:14:08,560
with the underlying states--
for example, hunger.

1577
01:14:08,560 --> 01:14:10,570
Food is tastier
when you're hungry.

1578
01:14:10,570 --> 01:14:14,740
Going on dates is less
enjoyable while being sick.

1579
01:14:14,740 --> 01:14:17,520
Classes are best
when you're rested.

1580
01:14:17,520 --> 01:14:20,842
Now, people know that
people's preferences

1581
01:14:20,842 --> 01:14:21,800
vary with their states.

1582
01:14:21,800 --> 01:14:22,675
They understand that.

1583
01:14:22,675 --> 01:14:25,750
If you ask hungry people about
are your preferences different

1584
01:14:25,750 --> 01:14:27,820
when you're hungry
versus not, most people

1585
01:14:27,820 --> 01:14:30,070
would probably say they want
different types of kinds

1586
01:14:30,070 --> 01:14:32,140
of food and the like and
they behave differently

1587
01:14:32,140 --> 01:14:34,240
when they're hungry.

1588
01:14:34,240 --> 01:14:38,990
However, there are biases
in state-dependent decision

1589
01:14:38,990 --> 01:14:39,490
making.

1590
01:14:39,490 --> 01:14:42,130
In particular, both
intuition and psychology

1591
01:14:42,130 --> 01:14:43,960
suggests that we
fail to appreciate

1592
01:14:43,960 --> 01:14:46,120
the extent to which
our preferences change

1593
01:14:46,120 --> 01:14:47,950
with the underlying states.

1594
01:14:47,950 --> 01:14:50,740
Projection bias is a
specific psychological error

1595
01:14:50,740 --> 01:14:51,850
of this type.

1596
01:14:51,850 --> 01:14:55,150
People overestimate the extent
to which future tastes resemble

1597
01:14:55,150 --> 01:14:57,880
their current tastes, and they
underestimate the influence

1598
01:14:57,880 --> 01:14:59,860
that the state has
on their utility,

1599
01:14:59,860 --> 01:15:02,590
and that can lead to
systematic misprediction.

1600
01:15:02,590 --> 01:15:05,170
I showed you quite
a few applications.

1601
01:15:05,170 --> 01:15:07,360
In particular,
addiction and depression

1602
01:15:07,360 --> 01:15:10,390
might be the most
consequential applications

1603
01:15:10,390 --> 01:15:12,960
of projection bias.

1604
01:15:12,960 --> 01:15:15,833
Now there's one more
bias that I told you

1605
01:15:15,833 --> 01:15:18,000
I'm going to talk about
very briefly-- again, that's

1606
01:15:18,000 --> 01:15:20,310
going to be discussed in
detail in recitation-- which

1607
01:15:20,310 --> 01:15:22,800
is called attribution bias.

1608
01:15:22,800 --> 01:15:26,310
Now attribution bias in
some sense is very similar,

1609
01:15:26,310 --> 01:15:28,500
but it's backward looking.

1610
01:15:28,500 --> 01:15:31,920
So what projection bias is the
misprediction of the influence

1611
01:15:31,920 --> 01:15:33,400
of future states.

1612
01:15:33,400 --> 01:15:36,008
So when you think about the
future, are you going to be--

1613
01:15:36,008 --> 01:15:37,800
how are you going to
enjoy a meal depending

1614
01:15:37,800 --> 01:15:41,610
on when you're hungry versus
not or burgers versus salad?

1615
01:15:41,610 --> 01:15:46,440
How is your future preferences
shaped by future states?

1616
01:15:46,440 --> 01:15:49,710
Attribution bias is
instead backward looking.

1617
01:15:49,710 --> 01:15:54,130
It's the misprediction of
the influence of past states.

1618
01:15:54,130 --> 01:15:56,460
So that's to say-- and I'm
going to sort of define that

1619
01:15:56,460 --> 01:15:57,480
briefly--

1620
01:15:57,480 --> 01:15:59,190
when judging the
value of a good,

1621
01:15:59,190 --> 01:16:02,550
people are overly influenced
by the state in which they

1622
01:16:02,550 --> 01:16:05,523
previously consumed it.

1623
01:16:05,523 --> 01:16:06,940
Now let me give
you some examples.

1624
01:16:06,940 --> 01:16:10,590
And again, then you'll talk
about this more in recitation.

1625
01:16:10,590 --> 01:16:13,170
People are more likely to
return to a restaurant when

1626
01:16:13,170 --> 01:16:14,328
first trying when hungry.

1627
01:16:14,328 --> 01:16:16,620
So if you go to a restaurant
when you're really, really

1628
01:16:16,620 --> 01:16:19,562
hungry, you'll think
the food is amazing

1629
01:16:19,562 --> 01:16:22,020
while, in fact, it might just
be that you're really, really

1630
01:16:22,020 --> 01:16:22,510
hungry.

1631
01:16:22,510 --> 01:16:24,385
So when you then come
back to the restaurant,

1632
01:16:24,385 --> 01:16:28,200
you might be surprised that,
in fact, it's not that great.

1633
01:16:28,200 --> 01:16:32,610
People are also more likely to
negatively rate a movie when

1634
01:16:32,610 --> 01:16:33,870
they've seen it while tired.

1635
01:16:33,870 --> 01:16:34,890
Well, why is that?

1636
01:16:34,890 --> 01:16:37,017
Because the experience
is just not that great,

1637
01:16:37,017 --> 01:16:39,600
and then you sort of think the
movie is really not that great.

1638
01:16:39,600 --> 01:16:41,740
But, really, it's
just you were tired.

1639
01:16:41,740 --> 01:16:44,550
People are also less likely
to recommend a zoo to a friend

1640
01:16:44,550 --> 01:16:47,220
if it rained during
the last visit.

1641
01:16:47,220 --> 01:16:50,310
Well, it's just the experience
of that zoo is not that great.

1642
01:16:50,310 --> 01:16:53,590
Of course, some days it rains
and some days it doesn't.

1643
01:16:53,590 --> 01:16:56,040
So you're not going to tell
usually your friend to only go

1644
01:16:56,040 --> 01:16:57,220
there when it's sunny.

1645
01:16:57,220 --> 01:17:00,610
People are saying the zoo
was just not that much fun.

1646
01:17:00,610 --> 01:17:03,040
And perhaps more
relevant for you,

1647
01:17:03,040 --> 01:17:04,320
people are more likely to--

1648
01:17:04,320 --> 01:17:07,890
or may be more likely to
recommend a class that they

1649
01:17:07,890 --> 01:17:10,440
took while well rested.

1650
01:17:10,440 --> 01:17:15,600
So a few years ago I was
teaching 1413 at 9:00 AM,

1651
01:17:15,600 --> 01:17:19,560
and I got lots of complaints
from students about the class

1652
01:17:19,560 --> 01:17:21,000
being way too early.

1653
01:17:21,000 --> 01:17:25,470
And I was telling them or trying
to tell them that the class is

1654
01:17:25,470 --> 01:17:27,450
really lots of fun.

1655
01:17:27,450 --> 01:17:31,690
You may or may not be able to
appreciate that if you really

1656
01:17:31,690 --> 01:17:32,490
are tired.

1657
01:17:32,490 --> 01:17:36,330
So attribution bias might
sort make you think that--

1658
01:17:36,330 --> 01:17:40,140
might attribute some negative
or positive experience

1659
01:17:40,140 --> 01:17:44,430
to the actual underlying
quality of the issue

1660
01:17:44,430 --> 01:17:48,030
or the good that's being offered
as opposed to, as you should,

1661
01:17:48,030 --> 01:17:49,650
to the underlying state.

1662
01:17:49,650 --> 01:17:54,000
Again, you're going to talk
more about this in recitation.

1663
01:17:54,000 --> 01:17:55,290
Now what's next?

1664
01:17:55,290 --> 01:17:58,660
Remember, there is no
class on Monday, April 20.

1665
01:17:58,660 --> 01:18:00,720
On Wednesday, April
22, we're going

1666
01:18:00,720 --> 01:18:03,390
to talk about gender,
discrimination, and identity.

1667
01:18:03,390 --> 01:18:06,710
Please read Heather
Sarson's paper, section one,

1668
01:18:06,710 --> 01:18:08,680
just the introduction.

1669
01:18:08,680 --> 01:18:11,270
Thank you very much.